Bioinformatics  Programming  2013

Problem  1
Installing  Python

Why  Python?

Rosalind  problems  can  be  solved  using  any  programming  language.  Our  language
of  choice  is  Python.  Why?  Because  it's  simple,  powerful,  and  even  funny.  You'll
see  what  we  mean.

If  you  don't  already  have  Python  software,  please  download  and  install  the  appropriate  version  for
your  platform  (Windows,  Linux  or  Mac  OS  X).  Please  install  Python  of  version  2.x  (not  3.x)  —  it  has
more  libraries  support  and  many  well-­written  guides.

After  completing  installation,  launch  IDLE  (default  Python  development  environment;;  it's  usually
installed  with  Python,  however  you  may  need  to  install  it  separately  on  Linux).

You'll  see  a  window  containing  three  arrows,  like  so:

>>>

The  three  arrows  are  Python's  way  of  saying  that  it  is  ready  to  serve  your  every  need.  You  are  in
interactive  mode,  meaning  that  any  command  you  type  will  run  immediately.  Try  typing   1+1  and
see  what  happens.

Of  course,  to  become  a  Rosalind  pro,  you  will  need  to  write  programs  having  more  than  one  line.
So  select  File  →  New  Window  from  the  IDLE  menu.  You  can  now  type  code  as  you  would  into  a
text  editor.  For  example,  type  the  following:

print  "Hello,  World!"

Select  File  →  Save  to  save  your  creation  with  an  appropriate  name  (e.g.,   hello.py ).

To  run  your  program,  select  Run  →  Run  Module.  You'll  see  the  result  in  the  interactive  mode
window  (Python  Shell).

Congratulations!  You  just  ran  your  first  program  in  Python!

Problem

After  downloading  and  installing  Python,  type   import  this  into  the  Python  command  line  and  see
what  happens.  Then,  click  the  "Download  dataset"  button  below  and  copy  the  Zen  of  Python  into  the
space  provided.

Problem  2
Variables  and  Some  Arithmetic

Variables  and  Some  Arithmetic

One  of  the  most  important  features  of  any  programming  language  is  its  ability  to
manipulate  variables.  A  variable  is  just  a  name  that  refers  to  a  value;;  you  can
think  of  a  variable  as  a  box  that  stores  a  piece  of  data.

In  Python,  the  basic  data  types  are  strings  and  numbers.  There  are  two  types  of  numbers:  integers
(both  positive  and  negative)  and  floats  (fractional  numbers  with  a  decimal  point).  You  can  assign
numbers  to  variables  very  easily.  Try  running  the  following  program:

a  =  324
b  =  24
c  =  a  -­‐  b
print  'a  -­‐  b  is',  c

In  the  above  code,  a,  b,  and  c  are  all  integers,  and  'a  -­  b  is'  is  a  string.  The  result  of  this  program  is
to  print:

a  -­‐  b  is  300

You  can  now  use  all  common  arithmetic  operations  involving  numbers:

Addition:   2  +  3  ==  5
Subtraction:   5  -­‐  2  ==  3
Multiplication:   3  *  4  ==  12
Division:   15  /  3  ==  5
Division  remainder:   18  %  5  ==  3
Exponentiation:   2  **  3  ==  8

It  is  important  to  note  that  if  you  try  to  divide  two  integers,  Python  always  rounds  down  the  result
(so   18/5  ==  3 ).

To  obtain  a  precise  result  for  this  division,  you  need  to  indicate  floating  point  division;;  either  of  the
following  expressions  results  in  a  "float"  data  type:   18.0/5  ==  3.6  or   float(18)/5  ==  3.6

In  Python,  the  single  equals  sign  ( = )  means  "assign  a  value  to  a  variable".  For  example,   a  =  3
assigns  3  to  the  integer  a.  In  order  to  denote  equality,  Python  uses  the  double  equals  sign  ( == ).

In  Python,  a  string  is  an  ordered  sequence  of  letters,  numbers  and  other  characters.  You  can
create  string  variables  just  like  you  did  with  :

a  =  "Hello"
b  =  "World"

Notice  that  the  string  must  be  surrounded  by  "  or  '  (but  not  a  mix  of  both).  You  can  use  quotes
inside  the  string,  as  long  as  you  use  the  opposite  type  of  quotes  to  surround  the  string,  e.g.,   a  =
"Monty  Python's  Flying  Circus"  or   b  =  'Project  "Rosalind"' .

String  operations  differ  slightly  from  operations  on  numbers:

a  =  'Rosalind'
b  =  'Franklin'
 c  =  '!'
print  a  +  '  '  +  b  +  c*3

Output:

Rosalind  Franklin!!!

Problem

Given:  Two  positive  integers    and   ,  each  less  than  1000.

Return:  The  integer  corresponding  to  the  square  of  the  hypotenuse  of  the  right  triangle  whose  legs
have  lengths    and   .

Notes:

1.   The  dataset  changes  every  time  you  click  "Download  dataset".

2.   We  check  only  your  final  answer  to  the  downloaded  dataset  in  the  box  below,  not  your  code  itself.
If  you  would  like  to  provide  your  code  as  well,  you  may  use  the  upload  tool.  Please  also  note  that
the  correct  answer  to  this  problem  will  not  in  general  be  34;;  it  is  simply  an  example  of  what  you
should  return  in  the  specific  case  that  the  legs  of  the  triangle  have  length  3  and  5.

Sample  Dataset

3  5

Sample  Output

34

Problem  3
Strings  and  Lists

Strings  and  lists

We've  already  seen  numbers  and  strings,  but  Python  also  has  variable  types  that
can  hold  more  than  one  piece  of  data  at  a  time.  The  simplest  such  variable  is  a
list.

You  can  assign  data  to  a  list  in  the  following  way:   list_name  =  [item_1,  item_2,  ...,
item_n] .  The  items  of  the  list  can  be  of  any  other  type:  integer,  float,  string.  You  even  explore  your
inner  Zen  and  make  lists  of  lists!

Any  item  in  a  list  can  be  accessed  by  its  index,  or  the  number  that  indicates  its  place  in  the  list.
For  example,  try  running  the  following  code:
 tea_party  =  ['March  Hare',  'Hatter',  'Dormouse',  'Alice']
print  tea_party[2]

Your  output  should  be:

Dormouse

Note  that  the  output  was  not   Hatter ,  as  you  might  have  guessed.  This  is  because  in  Python,
indexing  begins  with  0,  not  1.  This  property  is  called  0-­based  numbering,  and  it's  shared  by  many
programming  languages.

You  can  easily  change  existing  list  items  by  reassigning  them.  Try  running  the  following:

tea_party[1]  =  'Cheshire  Cat'

print  tea_party

This  code  should  output  the  list  with  "Hatter"  replaced  by  "Cheshire  Cat":

March  Hare,  Cheshire  Cat,  Dormouse,  Alice

You  can  also  add  items  to  the  end  of  an  existing  list  by  using  the  function   append() :

tea_party.append('Jabberwocky')
print  tea_party

This  code  outputs  the  following:

March  Hare,  Cheshire  Cat,  Dormouse,  Alice,  Jabberwocky

If  you  need  to  obtain  only  some  of  a  list,  you  can  use  the  notation   list_name[a:b]  to  get  only
those  from  index   a  up  to  but  not  including  index   b .  For  example,   tea_party[1:3]  returns
Cheshire  Cat,  Dormouse ,  not   Cheshire  Cat,  Dormouse,  Alice .  This  process  is  called  "list
slicing".

If  the  first  index  of  the  slice  is  unspecified,  then  Python  assumes  that  the  slice  begins  with  the
beginning  of  the  list  (i.e.,  index  0);;  if  the  second  index  of  the  slice  is  unspecified,  then  you  will
obtain  the  items  at  the  end  of  the  list.  For  example,   tea_party[:2]  returns   March  Hare,
Cheshire  Cat  and   tea_party[3:]  returns   Alice,  Jabberwocky .

You  can  also  use  negative  indices  to  count  items  backtracking  from  the  end  of  the  list.  So
tea_party[-­‐2:]  returns  the  same  output  as   tea_party[3:] :   Alice,  Jabberwocky .

Finally,  Python  equips  you  with  the  magic  ability  to  slice  strings  the  same  way  that  you  slice  lists.
A  string  can  be  considered  as  a  list  of  characters,  each  of  which  having  its  own  index  starting  from
0.  For  example,  try  running  the  following  code:

a  =  'flimsy'
b  =  'miserable'
c  =  b[0:1]  +  a[2:]
print  c

This  code  will  output  the  string  formed  by  the  first  character  of   miserable  and  the  last  four
characters  of   flimsy :
 mimsy

Problem

Given:  A  string    of  length  at  most  200  letters  and  four  integers   ,   ,    and   .
Return:  The  slice  of  this  string  from  indices    through    and    through    (with  space  in  between),
inclusively.

Sample  Dataset

HumptyDumptysatonawallHumptyDumptyhadagreatfallAlltheKingshorsesandallt
heKingsmenCouldntputHumptyDumptyinhisplaceagain.
22  27  97  102

Sample  Output

Humpty  Dumpty

Problem  4
Conditions  and  Loops

Conditions  and  Loops

If  you  need  Python  to  choose  between  two  actions,  then  you  can  use  an
if / else  statement.  Try  running  this  example  code:

a  =  42
if  a  <  10:
   print  'the  number  is  less  than  10'
else:
   print  'the  number  is  greater  or  equal  to  10'

Note  the  indentation  and  punctuation  (especially  the  colons),  because  they  are  important.

If  we  leave  out  an   else ,  then  the  program  continues  on.  Try  running  this  program  with  different
initial  values  of   a  and   b :

if  a  +  b  ==  4:
   print  'printed  when  a  +  b  equals  four'
print  'always  printed'

If  you  want  to  repeat  an  action  several  times,  you  can  use  a  while  loop.  The  following  program
 prints   Hello  once,  then  adds  1  to  the   greetings  counter.  It  then  prints   Hello  twice  because
greetings  is  equal  to  2,  then  adds  1  to   greetings .  After  printing   Hello  three  times,
greetings  becomes  4,  and  the   while  condition  of   greetings  <=  3  is  no  longer  satisfied,  so
the  program  would  continue  past  the   while  loop.

greetings  =  1
while  greetings  <=  3:
   print  'Hello!  '  *  greetings
   greetings  =  greetings  +  1

Be  careful!  If  you  accidentally  create  an  infinite  loop,  your  program  will  freeze  and  you  will  have  to
abort  it.  Here's  an  example  of  an  infinite  loop.  Make  sure  you  see  why  it  will  never  exit  the   while
loop:

greetings  =  1
while  greetings  <=  3:
   print  'Hello!  '  *  greetings
   greetings  =  greetings  +  0  #  Bug  here

If  you  want  to  carry  out  some  action  on  every  element  of  a  list,  the   for  loop  will  be  handy

names  =  ['Alice',  'Bob',  'Charley']

for  name  in  names:
   print  'Hello,  '  +  name

And  if  you  want  to  repeat  an  action  exactly    times,  you  can  use  the  following  template:

n  =  10
for  i  in  range(n):
   print  i

In  the  above  code,   range  is  a  function  that  creates  a  list  of  integers  between    and   ,  where    is
not  included.

Finally,  try  seeing  what  the  following  code  prints  when  you  run  it:

print  range(5,  12)

More  information  about  loops  and  conditions  can  be  found  in  the  Python  documentation.

Problem

Given:  Two  positive  integers    and    ( ).

Return:  The  sum  of  all  odd  integers  from    through   ,  inclusively.

Sample  Dataset

100  200

Sample  Output
 7500

Hint

You  can  use   a  %  2  ==  1  to  test  if  a  is  odd.

Problem  5
Working  with  Files

Reading  and  Writing

In  Rosalind,  sample  datasets  are  given  as  files.  Python  has  a  lot  of  functions  for
reading  and  writing  information  in  files.

To  access  a  file,  you  must  first  open  it.  To  do  so,  you  can  use  the   open()
function,  which  takes  two  parameters:  the  name  of  the  target  file  and  the  mode.  Three  modes  are
available:

r  -­  read  mode  (the  file  is  opened  for  reading)

w  -­  write  mode  (the  file  is  opened  for  writing,  and  if  a  file  having  the  same  name  exists,  it  will
be  erased)
a  -­  append  mode  (the  file  is  opened  for  appending,  which  means  that  data  is  only  to  be  added
to  the  existing  data  in  the  file)

f  =  open('input.txt',  'r')

This  code  told  Python  to  open  the  file   input.txt  in   r  mode  and  store  the  result  of  this  operation
in  a  file  object  called   f .

To  obtain  data  from  the  file  object  you  created,  you  can  apply  the  following  methods:

The  command   f.read(n)  returns    bytes  of  data  from  the  file  as  a  string.  When  the  size
parameter  is  omitted,  the  entire  contents  of  the  file  will  be  read  and  returned.

The  command   f.readline()  takes  a  single  line  from  the  file.  Every  line  (except  the  last  line  of
file)  terminates  in  a  newline  character  ( \n ).  To  remove  this  character  from  the  end  of  a  line  you
have  read,  use  the   .strip()  method.  Note  that  every  time  you  call   .readline()  it  takes  the
next  line  in  the  file.

The  command   f.readlines()  returns  a  list  containing  every  line  in  the  file.  If  you  need  to  obtain  a
particular  line,  you  can  use  a  list  item  index,  e.g.,   f.readlines()[2]  returns  the  third  line  of  the
file  object   f  (don't  forget  that  Python  utilizes  0-­based  numbering!)

An  alternative  way  to  read  lines  is  to  loop  over  the  file  object.

for  line  in  f:

   print  line
 Using  this  loop,  you  can  do  anything  you  need  with  every  line  in  the  file.

If  the  data  in  the  file  are  not  separated  by  new  lines  but  rather  by  whitespace,  commas,  or  any
other  delimeter,  then  all  three  commands  above  will  return  the  data  only  in  the  form  of  lines.  As  a
workaround,  you  can  use  the  command   line.split() .  It  uses  whitespace  in  addition  to   \n  as
delimeters  by  default,  and  runs  of  the  same  delimiter  are  regarded  as  a  single  separating  space.
For  example,

'Beautiful  is  better  than  ugly.\n'.split()  returns   ['Beautiful',  'is',  'better',

'than',  'ugly.']

You  can  even  specify  the  delimiter  as  a  parameter  of   line.split() :

'Explicit,  is  better,  than  implicit.'.split(",")  returns   ['Explicit',  '  is

better',  '  than  implicit.']

Another  convenient  command  for  file  parsing  is   .splitlines() .  It  returns  a  list  of  the  lines  in  the
string,  breaking  at  line  boundaries.  Line  breaks  are  not  included.

'Simple  is\nbetter  than\ncomplex.\n'.splitlines()  returns   ['Simple  is',  'better

than',  'complex.']

When  you  at  last  complete  all  your  calculations  and  obtain  a  result,  you  need  to  store  it
somewhere.  To  save  a  file,  output  the  desired  file  in  write  mode  (if  there  is  no  such  file,  it  will  be
created  automatically):

f  =  open('output.txt',  'w')

You  can  then  write  your  data  using   .write()  method.

f.write('Any  data  you  want  to  write  into  file')

The  command   f.write(string)  writes  the  contents  of   string  to  file   f .  If  you  want  to  write
something  other  than  a  string  (an  integer  say),  you  must  first  convert  it  to  a  string  by  using  the
function   str() .

inscription  =  ['Rosalind  Elsie  Franklin  ',  1920,  1958]

s  =  str(inscription)
f.write(s)

You  also  can  write  list  items  into  a  file  one  at  a  time  by  using  a   for  loop:

for  i  in  inscription:

   output.write(str(i)  +  '\n')

Adding   \n  to   str(i)  means  that  every  item  will  start  with  a  new  line.

When  you  are  finished  writing  file,  don't  forget  that  you  must  close  it  using  the  command
f.close() .  It's  a  good  habit  to  get  into.

Problem

Given:  A  file  containing  at  most  1000  lines.

Return:  A  file  containing  all  the  even-­numbered  lines  from  the  original  file.  Assume  1-­based
numbering  of  lines.
 Sample  Dataset

`Bravely  bold  Sir  Robin  rode  forth  from  Camelot

Yes,  brave  Sir  Robin  turned  about
He  was  not  afraid  to  die,  O  brave  Sir  Robin
And  gallantly  he  chickened  out
He  was  not  at  all  afraid  to  be  killed  in  nasty  ways
Bravely  talking  to  his  feet
Brave,  brave,  brave,  brave  Sir  Robin
He  beat  a  very  brave  retreat

Sample  Output

Yes,  brave  Sir  Robin  turned  about

And  gallantly  he  chickened  out
Bravely  talking  to  his  feet
He  beat  a  very  brave  retreat

Problem  6
Dictionaries

Intro  to  Python  dictionary

We've  already  used  lists  and  strings  to  store  and  process  bunch  of  data.  Python
also  has  a  variable  type  to  matching  one  items  (i.e.  keys)  to  others  (i.e.  values)
called  dictionary.  Dictionary  is  similar  to  list  but  instead  of  automatic  index  you
provide  your  own  index  called  key.  You  can  assign  data  to  a  dictionary  as  follows:
phones  =  {'Zoe':'232-­‐43-­‐58',  'Alice':'165-­‐88-­‐56'} .  As  with  lists  for  a  value  could  be
used  any  type:  string,  number,  float  even  dict  or  list.  For  keys  you  can  use  only  strings,  numbers,
floats  and  other  immutable  types.  Accessing  values  also  similar  to  lists:

phones  =  {'Zoe':'232-­‐43-­‐58',  'Alice':'165-­‐88-­‐56'}

print  phones['Zoe']

Output  should  be:

232-­‐43-­‐58

Adding  new  value  to  dictionary  or  assigning  an  existent  can  be  done  the  same  way  as  you  do  it
with  variable

phones['Zoe']  =  '658-­‐99-­‐55'
phones['Bill']  =  '342-­‐18-­‐25'
print  phones
 You  should  see  the  following:

{'Bill':  '342-­‐18-­‐25',  'Zoe':  '658-­‐99-­‐55',  'Alice':  '165-­‐88-­‐56'}

Note  that  new   'Bill'  appeared  in  the  beginning  not  in  the  end  as  you  might  expected.  The  thing
is  that  dictionary  basically  does  not  have  any  ordering.  New  value  appear  in  random  place.

Remember  that  the  dictionary  is  case-­sensitive  if  you  are  using  strings  as  keys.  Keep  in  mind  that
'key'  and  'Key'  are  different  keys:

d  =  {}
d['key']  =  1
d['Key']  =  2
d['KEY']  =  3
print  d

Output:

{'KEY':  3,  'Key':  2,  'key':  1}

Note  how  we  created  an  empty  dictionary  with   d  =  {} .  This  could  be  useful  in  case  you  need  to
add  values  to  dictionary  dynamically  (for  example,  when  reading  a  file).  If  you  need  to  check  is
there  a  key  in  dictionary  you  can  use   key  in  d  syntax:

if  'Peter'  in  phones:

       print  "We  know  Peter's  phone"
else:
       print  "We  don't  know  Peter's  phone"

Output:

We  don't  know  Peter's  phone

In  case  you  need  to  delete  a  value  from  a  dictionary  please  use   del :

phones  =  {'Zoe':'232-­‐43-­‐58',  'Alice':'165-­‐88-­‐56'}

del  phones['Zoe']
print  phones

Output:

{'Alice':  '165-­‐88-­‐56'}

Problem

Given:  A  string    of  length  at  most  10000  letters.

Return:  How  many  times  any  word  occurred  in  string.  Each  letter  case  (upper  or  lower)  in  word
matters.  Lines  in  output  can  be  in  any  order.

Sample  Dataset
 We  tried  list  and  we  tried  dicts  also  we  tried  Zen

Sample  Output

and  1
We  1
tried  3
dicts  1
list  1
we  2
also  1
Zen  1

Hints

To  iterate  over  words  in  string  you  can  split  it  by  space:

for  word  in  str.split('  '):

       print  word

To  have  nice  output  of  dictionary  you  can  use   .items()  method:

for  key,  value  in  dict.items():

       print  key
       print  value

Problem  7
Counting  DNA  Nucleotides

A  Rapid  Introduction  to  Molecular  Biology

Making  up  all  living  material,  the  cell  is

considered  to  be  the  building  block  of  life.
The  nucleus,  a  component  of  most
eukaryotic  cells,  was  identified  as  the
hub  of  cellular  activity  150  years  ago.  Viewed  under  a  light
microscope,  the  nucleus  appears  only  as  a  darker  region
of  the  cell,  but  as  we  increase  magnification,  we  find  that
the  nucleus  is  densely  filled  with  a  stew  of
macromolecules  called  chromatin.  During  mitosis
Figure  1.  A  1900  drawing  by
(eukaryotic  cell  division),  most  of  the  chromatin  condenses Edmund  Wilson  of  onion  cells  at
into  long,  thin  strings  called  chromosomes.  See  Figure  1 different  stages  of  mitosis.  The
sample  has  been  dyed,  causing
 for  a  figure  of  cells  in  different  stages  of  mitosis. chromatin  in  the  cells  (which  soaks
up  the  dye)  to  appear  in  greater
One  class  of  the  macromolecules  contained  in  chromatin contrast  to  the  rest  of  the  cell.
are  called  nucleic  acids.  Early  20th  century  research  into
the  chemical  identity  of  nucleic  acids  culminated  with  the
conclusion  that  nucleic  acids  are  polymers,  or  repeating
chains  of  smaller,  similarly  structured  molecules  known  as
monomers.  Because  of  their  tendency  to  be  long  and  thin,
nucleic  acid  polymers  are  commonly  called  strands.

The  nucleic  acid  monomer  is  called  a  nucleotide  and  is

used  as  a  unit  of  strand  length  (abbreviated  to  nt).  Each
nucleotide  is  formed  of  three  parts:  a  sugar  molecule,  a
negatively  charged  ion  called  a  phosphate,  and  a
compound  called  a  nucleobase  ("base"  for  short).
Polymerization  is  achieved  as  the  sugar  of  one  nucleotide
bonds  to  the  phosphate  of  the  next  nucleotide  in  the  chain, Figure  2.  A  sketch  of  DNA's  primary
which  forms  a  sugar-­phosphate  backbone  for  the  nucleic structure.
acid  strand.  A  key  point  is  that  the  nucleotides  of  a
specific  type  of  nucleic  acid  always  contain  the  same  sugar  and  phosphate  molecules,  and  they
differ  only  in  their  choice  of  base.  Thus,  one  strand  of  a  nucleic  acid  can  be  differentiated  from
another  based  solely  on  the  order  of  its  bases;;  this  ordering  of  bases  defines  a  nucleic  acid's
primary  structure.

For  example,  Figure  2  shows  a  strand  of  deoxyribose  nucleic  acid  (DNA),  in  which  the  sugar  is
called  deoxyribose,  and  the  only  four  choices  for  nucleobases  are  molecules  called  adenine  (A),
cytosine  (C),  guanine  (G),  and  thymine  (T).

For  reasons  we  will  soon  see,  DNA  is  found  in  all  living  organisms  on  Earth,  including  bacteria;;  it  is
even  found  in  many  viruses  (which  are  often  considered  to  be  nonliving).  Because  of  its  importance,
we  reserve  the  term  genome  to  refer  to  the  sum  total  of  the  DNA  contained  in  an  organism's
chromosomes.

Problem

A  string  is  simply  an  ordered  collection  of  symbols  selected  from  some  alphabet  and  formed  into  a
word;;  the  length  of  a  string  is  the  number  of  symbols  that  it  contains.

An  example  of  a  length  21  DNA  string  (whose  alphabet  contains  the  symbols  'A',  'C',  'G',  and  'T')  is
"ATGCTTCAGAAAGGTCTTACG."

Given:  A  DNA  string    of  length  at  most  1000  nt.

Return:  Four  integers  (separated  by  spaces)  counting  the  respective  number  of  times  that  the
symbols  'A',  'C',  'G',  and  'T'  occur  in   .

Sample  Dataset

AGCTTTTCATTCTGACTGCAACGGGCAATATGTCTCTGTGTGGATTAAAAAAAGAGTGTCTGATAGCAGC

Sample  Output

20  12  17  21

Problem  8
Transcribing  DNA  into  RNA

The  Second  Nucleic  Acid

In  “Counting  DNA  Nucleotides”,  we

described  the  primary  structure  of  a
nucleic  acid  as  a  polymer  of  nucleotide
units,  and  we  mentioned  that  the
omnipresent  nucleic  acid  DNA  is  composed  of  a  varied
sequence  of  four  bases.

Yet  a  second  nucleic  acid  exists  alongside  DNA  in  the

chromatin;;  this  molecule,  which  possesses  a  different
sugar  called  ribose,  came  to  be  known  as  ribose  nucleic
acid,  or  RNA.  RNA  differs  further  from  DNA  in  that  it
contains  a  base  called  uracil  in  place  of  thymine;;
structural  differences  between  DNA  and  RNA  are  shown  in Figure  1.  Structural  differences
Figure  1.  Biologists  initially  believed  that  RNA  was  only between  RNA  and  DNA
contained  in  plant  cells,  whereas  DNA  was  restricted  to
animal  cells.  However,  this  hypothesis  dissipated  as  improved  chemical  methods  discovered  both
nucleic  acids  in  the  cells  of  all  life  forms  on  Earth.

The  primary  structure  of  DNA  and  RNA  is  so  similar  because  the  former  serves  as  a  blueprint  for
the  creation  of  a  special  kind  of  RNA  molecule  called  messenger  RNA,  or  mRNA.  mRNA  is
created  during  RNA  transcription,  during  which  a  strand  of  DNA  is  used  as  a  template  for
constructing  a  strand  of  RNA  by  copying  nucleotides  one  at  a  time,  where  uracil  is  used  in  place  of
thymine.

In  eukaryotes,  DNA  remains  in  the  nucleus,  while  RNA  can  enter  the  far  reaches  of  the  cell  to  carry
out  DNA's  instructions.  In  future  problems,  we  will  examine  the  process  and  ramifications  of  RNA
transcription  in  more  detail.

Problem

An  RNA  string  is  a  string  formed  from  the  alphabet  containing  'A',  'C',  'G',  and  'U'.

Given  a  DNA  string    corresponding  to  a  coding  strand,  its  transcribed  RNA  string    is  formed  by
replacing  all  occurrences  of  'T'  in    with  'U'  in   .

Given:  A  DNA  string    having  length  at  most  1000  nt.

Return:  The  transcribed  RNA  string  of   .

Sample  Dataset

GATGGAACTTGACTACGTAAATT
 Sample  Output

GAUGGAACUUGACUACGUAAAUU

Problem  9
Complementing  a  Strand  of  DNA

The  Secondary  and  Tertiary  Structures  of  DNA

In  “Counting  DNA  Nucleotides”,  we

introduced  nucleic  acids,  and  we  saw
that  the  primary  structure  of  a  nucleic
acid  is  determined  by  the  ordering  of  its
nucleobases  along  the  sugar-­phosphate  backbone  that
constitutes  the  bonds  of  the  nucleic  acid  polymer.  Yet
primary  structure  tells  us  nothing  about  the  larger,  3-­
dimensional  shape  of  the  molecule,  which  is  vital  for  a
complete  understanding  of  nucleic  acids.

The  search  for  a  complete  chemical  structure  of  nucleic

acids  was  central  to  molecular  biology  research  in  the  mid-­
20th  Century,  culminating  in  1953  with  a  publication  in Figure  1.  Base  pairing  across  the
Nature  of  fewer  than  800  words  by  James  Watson  and two  strands  of  DNA.
Francis  Crick.  Consolidating  a  high  resolution  X-­ray  image
created  by  Rosalind  Franklin  and  Raymond  Gosling  with  a
number  of  established  chemical  results,  Watson  and  Crick
proposed  the  following  structure  for  DNA:

1.   The  DNA  molecule  is  made  up  of  two  strands,  running
in  opposite  directions.
2.   Each  base  bonds  to  a  base  in  the  opposite  strand.
Adenine  always  bonds  with  thymine,  and  cytosine
always  bonds  with  guanine;;  the  complement  of  a
base  is  the  base  to  which  it  always  bonds;;  see  Figure
1.
3.   The  two  strands  are  twisted  together  into  a  long  spiral
staircase  structure  called  a  double  helix;;  see  Figure Figure  2.  The  double  helix  of  DNA  on
the  molecular  scale.
2.

Because  they  dictate  how  bases  from  different  strands  interact  with  each  other,  (1)  and  (2)  above
compose  the  secondary  structure  of  DNA.  (3)  describes  the  3-­dimensional  shape  of  the  DNA
molecule,  or  its  tertiary  structure.

In  light  of  Watson  and  Crick's  model,  the  bonding  of  two  complementary  bases  is  called  a  base
pair  (bp).  Therefore,  the  length  of  a  DNA  molecule  will  commonly  be  given  in  bp  instead  of  nt.  By
complementarity,  once  we  know  the  order  of  bases  on  one  strand,  we  can  immediately  deduce  the
sequence  of  bases  in  the  complementary  strand.  These  bases  will  run  in  the  opposite  order  to
match  the  fact  that  the  two  strands  of  DNA  run  in  opposite  directions.
 Problem

In  DNA  strings,  symbols  'A'  and  'T'  are  complements  of  each  other,  as  are  'C'  and  'G'.

The  reverse  complement  of  a  DNA  string    is  the  string    formed  by  reversing  the  symbols  of   ,
then  taking  the  complement  of  each  symbol  (e.g.,  the  reverse  complement  of  "GTCA"  is  "TGAC").

Given:  A  DNA  string    of  length  at  most  1000  bp.

Return:  The  reverse  complement    of   .

Sample  Dataset

AAAACCCGGT

Sample  Output

ACCGGGTTTT

Problem  10
Rabbits  and  Recurrence  Relations

Wascally  Wabbits

In  1202,  Leonardo  of  Pisa  (commonly

known  as  Fibonacci)  considered  a
mathematical  exercise  regarding  the
reproduction  of  a  population  of  rabbits.  He
made  the  following  simplifying  assumptions  about  the
population:

1.   The  population  begins  in  the  first  month  with  a  pair  of
newborn  rabbits.
2.   Rabbits  reach  reproductive  age  after  one  month.
3.   In  any  given  month,  every  rabbit  of  reproductive  age
mates  with  another  rabbit  of  reproductive  age. Figure  1.  The  growth  of  Fibonacci's
rabbit  population  for  the  first  six
4.   Exactly  one  month  after  two  rabbits  mate,  they months.
produce  one  male  and  one  female  rabbit.
5.   Rabbits  never  die  or  stop  reproducing.

Fibonacci's  exercise  was  to  calculate  how  many  pairs  of  rabbits  would  remain  in  one  year.  We  can
see  that  in  the  second  month,  the  first  pair  of  rabbits  reach  reproductive  age  and  mate.  In  the  third
month,  another  pair  of  rabbits  is  born,  and  we  have  two  rabbit  pairs;;  our  first  pair  of  rabbits  mates
again.  In  the  fourth  month,  another  pair  of  rabbits  is  born  to  the  original  pair,  while  the  second  pair
reach  maturity  and  mate  (with  three  total  pairs).  The  dynamics  of  the  rabbit  population  are
illustrated  in  Figure  1.  After  a  year,  the  rabbit  population  boasts  144  pairs.
 Although  Fibonacci's  assumption  of  the  rabbits'  immortality
may  seem  a  bit  farfetched,  his  model  was  not  unrealistic
for  reproduction  in  a  predator-­free  environment:  European
rabbits  were  introduced  to  Australia  in  the  mid  19th
Century,  a  place  with  no  real  indigenous  predators  for
them.  Within  50  years,  the  rabbits  had  already  eradicated
many  plant  species  across  the  continent,  leading  to
irreversible  changes  in  the  Australian  ecosystem  and
turning  much  of  its  grasslands  into  eroded,  practically
uninhabitable  parts  of  the  modern  Outback  (see  Figure  2).
In  this  problem,  we  will  use  the  simple  idea  of  counting
rabbits  to  introduce  a  new  computational  topic,  which
involves  building  up  large  solutions  from  smaller  ones. Figure  2.  Erosion  at  Lake  Mungo  in
New  South  Wales,  which  was
initiated  by  European  rabbits  in  the
19th  Century.  Courtesy  Pierre
Problem Pouliquin.

A  sequence  is  an  ordered  collection  of  objects  (usually  numbers),  which  are  allowed  to  repeat.
Sequences  can  be  finite  or  infinite.  Two  examples  are  the  finite  sequence    and  the
infinite  sequence  of  odd  numbers   .  We  use  the  notation    to  represent  the   -­th
term  of  a  sequence.

A  recurrence  relation  is  a  way  of  defining  the  terms  of  a  sequence  with  respect  to  the  values  of
previous  terms.  In  the  case  of  Fibonacci's  rabbits  from  the  introduction,  any  given  month  will  contain
the  rabbits  that  were  alive  the  previous  month,  plus  any  new  offspring.  A  key  observation  is  that  the
number  of  offspring  in  any  month  is  equal  to  the  number  of  rabbits  that  were  alive  two  months  prior.  As
a  result,  if    represents  the  number  of  rabbit  pairs  alive  after  the   -­th  month,  then  we  obtain  the
Fibonacci  sequence  having  terms    that  are  defined  by  the  recurrence  relation  
 (with    to  initiate  the  sequence).  Although  the  sequence  bears
Fibonacci's  name,  it  was  known  to  Indian  mathematicians  over  two  millennia  ago.

When  finding  the   -­th  term  of  a  sequence  defined  by  a  recurrence  relation,  we  can  simply  use  the
recurrence  relation  to  generate  terms  for  progressively  larger  values  of   .  This  problem  introduces  us
to  the  computational  technique  of  dynamic  programming,  which  successively  builds  up  solutions  by
using  the  answers  to  smaller  cases.

Given:  Positive  integers    and   .

Return:  The  total  number  of  rabbit  pairs  that  will  be  present  after    months  if  we  begin  with  1  pair
and  in  each  generation,  every  pair  of  reproduction-­age  rabbits  produces  a  litter  of    rabbit  pairs  (instead
of  only  1  pair).

Sample  Dataset

5  3

Sample  Output

19
Problem  11
Computing  GC  Content

Identifying  Unknown  DNA  Quickly

A  quick  method  used  by  early  computer

software  to  determine  the  language  of  a
given  piece  of  text  was  to  analyze  the
frequency  with  which  each  letter
appeared  in  the  text.  This  strategy  was  used  because
each  language  tends  to  exhibit  its  own  letter  frequencies,
and  as  long  as  the  text  under  consideration  is  long
enough,  software  will  correctly  recognize  the  language
quickly  and  with  a  very  low  error  rate.  See  Figure  1  for  a
table  compiling  English  letter  frequencies.
Figure  1.  The  table  above  was
You  may  ask:  what  in  the  world  does  this  linguistic computed  from  a  large  number  of
English  words  and  shows  for  any
problem  have  to  do  with  biology?  Although  two  members  of letter  the  frequency  with  which  it
the  same  species  will  have  different  genomes,  they  still appears  in  those  words.  These
share  the  vast  percentage  of  their  DNA;;  notably,  99.9%  of frequencies  can  be  used  to  reliably
identify  a  piece  of  English  text  and
the  3.2  billion  base  pairs  in  a  human  genome  are  common differentiate  it  from  that  of  another
to  almost  all  humans  (i.e.,  excluding  people  having  major language.  Taken  from
http://en.wikipedia.org/wiki/File:English_letter_freque
genetic  defects).  For  this  reason,  biologists  will  speak  of
the  human  genome,  meaning  an  average-­case  genome
derived  from  a  collection  of  individuals.  Such  an  average  case  genome  can  be  assembled  for  any
species,  a  challenge  that  we  will  soon  discuss.

The  biological  analog  of  identifying  unknown  text  arises  when  researchers  encounter  a  molecule  of
DNA  deriving  from  an  unknown  species.  Because  of  the  base  pairing  relations  of  the  two  DNA
strands,  cytosine  and  guanine  will  always  appear  in  equal  amounts  in  a  double-­stranded  DNA
molecule.  Thus,  to  analyze  the  symbol  frequencies  of  DNA  for  comparison  against  a  database,  we
compute  the  molecule's  GC-­content,  or  the  percentage  of  its  bases  that  are  either  cytosine  or
guanine.

In  practice,  the  GC-­content  of  most  eukaryotic  genomes  hovers  around  50%.  However,  because
genomes  are  so  long,  we  may  be  able  to  distinguish  species  based  on  very  small  discrepancies  in
GC-­content;;  furthermore,  most  prokaryotes  have  a  GC-­content  significantly  higher  than  50%,  so
that  GC-­content  can  be  used  to  quickly  differentiate  many  prokaryotes  and  eukaryotes  by  using
relatively  small  DNA  samples.

Problem

The  GC-­content  of  a  DNA  string  is  given  by  the  percentage  of  symbols  in  the  string  that  are  'C'  or  'G'.
For  example,  the  GC-­content  of  "AGCTATAG"  is  37.5%.  Note  that  the  reverse  complement  of  any
DNA  string  has  the  same  GC-­content.

DNA  strings  must  be  labeled  when  they  are  consolidated  into  a  database.  A  commonly  used  method
of  string  labeling  is  called  FASTA  format.  In  this  format,  the  string  is  introduced  by  a  line  that  begins
with  '>',  followed  by  some  labeling  information.  Subsequent  lines  contain  the  string  itself;;  the  first  line
to  begin  with  '>'  indicates  the  label  of  the  next  string.
 In  Rosalind's  implementation,  a  string  in  FASTA  format  will  be  labeled  by  the  ID  "Rosalind_xxxx",
where  "xxxx"  denotes  a  four-­digit  code  between  0000  and  9999.

Given:  At  most  10  DNA  strings  in  FASTA  format  (of  length  at  most  1  kbp  each).
Return:  The  ID  of  the  string  having  the  highest  GC-­content,  followed  by  the  GC-­content  of  that
string.  Rosalind  allows  for  a  default  error  of  0.001  in  all  decimal  answers  unless  otherwise  stated;;
please  see  the  note  on  absolute  error  below.

Sample  Dataset

>Rosalind_6404
CCTGCGGAAGATCGGCACTAGAATAGCCAGAACCGTTTCTCTGAGGCTTCCGGCCTTCCC
TCCCACTAATAATTCTGAGG
>Rosalind_5959
CCATCGGTAGCGCATCCTTAGTCCAATTAAGTCCCTATCCAGGCGCTCCGCCGAAGGTCT
ATATCCATTTGTCAGCAGACACGC
>Rosalind_0808
CCACCCTCGTGGTATGGCTAGGCATTCAGGAACCGGAGAACGCTTCAGACCAGCCCGGAC
TGGGAACCTGCGGGCAGTAGGTGGAAT

Sample  Output

Rosalind_0808
60.919540

Note  on  Absolute  Error

We  say  that  a  number    is  within  an  absolute  error  of    to  a  correct  solution  if    is  within    of  the
correct  solution.  For  example,  if  an  exact  solution  is  6.157892,  then  for    to  be  within  an  absolute
error  of  0.001,  we  must  have  that   ,  or   .

Error  bounding  is  a  vital  practical  tool  because  of  the  inherent  round-­off  error  in  representing
decimals  in  a  computer,  where  only  a  finite  number  of  decimal  places  are  allotted  to  any  number.
After  being  compounded  over  a  number  of  operations,  this  round-­off  error  can  become  evident.  As  a
result,  rather  than  testing  whether  two  numbers  are  equal  with   ,  you  may  wish  to  simply
verify  that    is  very  small.

The  mathematical  field  of  numerical  analysis  is  devoted  to  rigorously  studying  the  nature  of
computational  approximation.

Problem  12
Counting  Point  Mutations

Evolution  as  a  Sequence  of  Mistakes

 A  mutation  is  simply  a  mistake  that
occurs  during  the  creation  or  copying  of  a
nucleic  acid,  in  particular  DNA.  Because
nucleic  acids  are  vital  to  cellular
functions,  mutations  tend  to  cause  a
ripple  effect  throughout  the  cell.  Although
mutations  are  technically  mistakes,  a
very  rare  mutation  may  equip  the  cell  with  a  beneficial
attribute.  In  fact,  the  macro  effects  of  evolution  are
attributable  by  the  accumulated  result  of  beneficial
microscopic  mutations  over  many  generations.

The  simplest  and  most  common  type  of  nucleic  acid

mutation  is  a  point  mutation,  which  replaces  one  base Figure  1.  A  point  mutation  in  DNA
changing  a  C-­G  pair  to  an  A-­T  pair.
with  another  at  a  single  nucleotide.  In  the  case  of  DNA,  a
point  mutation  must  change  the  complementary  base
accordingly;;  see  Figure  1.

Two  DNA  strands  taken  from  different  organism  or  species  genomes  are  homologous  if  they  share
a  recent  ancestor;;  thus,  counting  the  number  of  bases  at  which  homologous  strands  differ  provides
us  with  the  minimum  number  of  point  mutations  that  could  have  occurred  on  the  evolutionary  path
between  the  two  strands.

We  are  interested  in  minimizing  the  number  of  (point)  mutations  separating  two  species  because  of
the  biological  principle  of  parsimony,  which  demands  that  evolutionary  histories  should  be  as
simply  explained  as  possible.

Problem

Given  two  strings    and    of  equal  length,  the  Hamming

distance  between    and   ,  denoted   ,  is  the  number
of  corresponding  symbols  that  differ  in    and   .  See  Figure  2.
Figure  2.  The  Hamming  distance
Given:  Two  DNA  strings    and    of  equal  length  (not between  these  two  strings  is  7.
Mismatched  symbols  are  colored
exceeding  1  kbp). red.

Return:  The  Hamming  distance   .

Sample  Dataset

GAGCCTACTAACGGGAT
CATCGTAATGACGGCCT

Sample  Output

Problem  13
Mendel's  First  Law

Introduction  to  Mendelian  Inheritance

Modern  laws  of  inheritance  were  first

described  by  Gregor  Mendel  (an
Augustinian  Friar)  in  1865.  The
contemporary  hereditary  model,  called
blending  inheritance,  stated  that  an  organism  must
exhibit  a  blend  of  its  parent's  traits.  This  rule  is  obviously
violated  both  empirically  (consider  the  huge  number  of
people  who  are  taller  than  both  their  parents)  and
statistically  (over  time,  blended  traits  would  simply  blend
into  the  average,  severely  limiting  variation).

Mendel,  working  with  thousands  of  pea  plants,  believed

that  rather  than  viewing  traits  as  continuous  processes, Figure  1.  A  Punnett  square
they  should  instead  be  divided  into  discrete  building  blocks representing  the  possible  outcomes
called  factors.  Furthermore,  he  proposed  that  every  factor of  crossing  a  heterozygous
organism  (Yy)  with  a  homozygous
possesses  distinct  forms,  called  alleles. recessive  organism  (yy);;  here,  the
dominant  allele  Y  corresponds  to
In  what  has  come  to  be  known  as  his  first  law  (also yellow  pea  pods,  and  the  recessive
known  as  the  law  of  segregation),  Mendel  stated  that  every allele  y  corresponds  to  green  pea
pods.
organism  possesses  a  pair  of  alleles  for  a  given  factor.  If  an
individual's  two  alleles  for  a  given  factor  are  the  same,  then
it  is  homozygous  for  the  factor;;  if  the  alleles  differ,  then  the  individual  is  heterozygous.  The  first
law  concludes  that  for  any  factor,  an  organism  randomly  passes  one  of  its  two  alleles  to  each
offspring,  so  that  an  individual  receives  one  allele  from  each  parent.

Mendel  also  believed  that  any  factor  corresponds  to  only  two  possible  alleles,  the  dominant  and
recessive  alleles.  An  organism  only  needs  to  possess  one  copy  of  the  dominant  allele  to  display
the  trait  represented  by  the  dominant  allele.  In  other  words,  the  only  way  that  an  organism  can
display  a  trait  encoded  by  a  recessive  allele  is  if  the  individual  is  homozygous  recessive  for  that
factor.

We  may  encode  the  dominant  allele  of  a  factor  by  a  capital  letter  (e.g.,   )  and  the  recessive  allele
by  a  lower  case  letter  (e.g.,   ).  Because  a  heterozygous  organism  can  possess  a  recessive  allele
without  displaying  the  recessive  form  of  the  trait,  we  henceforth  define  an  organism's  genotype  to
be  its  precise  genetic  makeup  and  its  phenotype  as  the  physical  manifestation  of  its  underlying
traits.

The  different  possibilities  describing  an  individual's  inheritance  of  two  alleles  from  its  parents  can  be
represented  by  a  Punnett  square;;  see  Figure  1  for  an  example.

Problem

Probability  is  the  mathematical  study  of  randomly  occurring

phenomena.  We  will  model  such  a  phenomenon  with  a
random  variable,  which  is  simply  a  variable  that  can  take  a
number  of  different  distinct  outcomes  depending  on  the  result
of  an  underlying  random  process.

For  example,  say  that  we  have  a  bag  containing  3  red  balls
 and  2  blue  balls.  If  we  let    represent  the  random  variable Figure  2.  The  probability  of  any
outcome  (leaf)  in  a  probability  tree
corresponding  to  the  color  of  a  drawn  ball,  then  the diagram  is  given  by  the  product  of
probability  of  each  of  the  two  outcomes  is  given  by   probabilities  from  the  start  of  the  tree
 and   . to  the  outcome.  For  example,  the
probability  that  X  is  blue  and  Y  is
blue  is  equal  to  (2/5)(1/4),  or  1/10.
Random  variables  can  be  combined  to  yield  new  random
variables.  Returning  to  the  ball  example,  let    model  the  color
of  a  second  ball  drawn  from  the  bag  (without  replacing  the  first  ball).  The  probability  of    being  red
depends  on  whether  the  first  ball  was  red  or  blue.  To  represent  all  outcomes  of    and   ,  we  therefore
use  a  probability  tree  diagram.  This  branching  diagram  represents  all  possible  individual
probabilities  for    and   ,  with  outcomes  at  the  endpoints  ("leaves")  of  the  tree.  The  probability  of  any
outcome  is  given  by  the  product  of  probabilities  along  the  path  from  the  beginning  of  the  tree;;  see
Figure  2  for  an  illustrative  example.

An  event  is  simply  a  collection  of  outcomes.  Because  outcomes  are  distinct,  the  probability  of  an
event  can  be  written  as  the  sum  of  the  probabilities  of  its  constituent  outcomes.  For  our  colored  ball
example,  let    be  the  event  "  is  blue."    is  equal  to  the  sum  of  the  probabilities  of  two  different
outcomes:   ,  or  
(see  Figure  2  above).

Given:  Three  positive  integers   ,   ,  and   ,  representing  a  population  containing  

organisms:    individuals  are  homozygous  dominant  for  a  factor,    are  heterozygous,  and    are
homozygous  recessive.

Return:  The  probability  that  two  randomly  selected  mating  organisms  will  produce  an  individual
possessing  a  dominant  allele  (and  thus  displaying  the  dominant  phenotype).  Assume  that  any  two
organisms  can  mate.

Sample  Dataset

2  2  2

Sample  Output

0.78333

Hint

Consider  simulating  inheritance  on  a  number  of  small  test  cases  in  order  to  check  your  solution.

Problem  14
Translating  RNA  into  Protein

The  Genetic  Code

 Just  as  nucleic  acids  are  polymers  of
nucleotides,  proteins  are  chains  of
smaller  molecules  called  amino  acids;;
20  amino  acids  commonly  appear  in
every  species.  Just  as  the  primary
structure  of  a  nucleic  acid  is  given  by  the
order  of  its  nucleotides,  the  primary
structure  of  a  protein  is  the  order  of  its  amino  acids.  Some
proteins  are  composed  of  several  subchains  called
polypeptides,  while  others  are  formed  of  a  single
polypeptide;;  see  Figure  1.

Proteins  power  every  practical  function  carried  out  by  the Figure  1.  The  human  hemoglobin
cell,  and  so  presumably,  the  key  to  understanding  life  lies molecule  consists  of  4  polypeptide
in  interpreting  the  relationship  between  a  chain  of  amino chains;;  α  subunits  are  shown  in  red
and  β  subunits  are  shown  in  blue
acids  and  the  function  of  the  protein  that  this  chain  of
amino  acids  eventually  constructs.  Proteomics  is  the  field
devoted  to  the  study  of  proteins.

How  are  proteins  created?  The  genetic  code,  discovered  throughout  the  course  of  a  number  of
ingenious  experiments  in  the  late  1950s,  details  the  translation  of  an  RNA  molecule  called
messenger  RNA  (mRNA)  into  amino  acids  for  protein  creation.  The  apparent  difficulty  in  translation
is  that  somehow  4  RNA  bases  must  be  translated  into  a  language  of  20  amino  acids;;  in  order  for
every  possible  amino  acid  to  be  created,  we  must  translate  3-­nucleobase  strings  (called  codons)
into  amino  acids.  Note  that  there  are    possible  codons,  so  that  multiple  codons  may
encode  the  same  amino  acid.  Two  special  types  of  codons  are  the  start  codon  (AUG),  which
codes  for  the  amino  acid  methionine  always  indicates  the  start  of  translation,  and  the  three  stop
codons  (UAA,  UAG,  UGA),  which  do  not  code  for  an  amino  acid  and  cause  translation  to  end.

The  notion  that  protein  is  always  created  from  RNA,  which  in  turn  is  always  created  from  DNA,
forms  the  central  dogma  of  molecular  biology.  Like  all  dogmas,  it  does  not  always  hold;;
however,  it  offers  an  excellent  approximation  of  the  truth.

A  eukaryotic  organelle  called  a  ribosome  creates  peptides  by  using  a  helper  molecule  called
transfer  RNA  (tRNA).  A  single  tRNA  molecule  possesses  a  string  of  three  RNA  nucleotides  on  one
end  (called  an  anticodon)  and  an  amino  acid  at  the  other  end.  The  ribosome  takes  an  RNA
molecule  transcribed  from  DNA  (see  “Transcribing  DNA  into  RNA”),  called  messenger  RNA
(mRNA),  and  examines  it  one  codon  at  a  time.  At  each  step,  the  tRNA  possessing  the
complementary  anticodon  bonds  to  the  mRNA  at  this  location,  and  the  amino  acid  found  on  the
opposite  end  of  the  tRNA  is  added  to  the  growing  peptide  chain  before  the  remaining  part  of  the
tRNA  is  ejected  into  the  cell,  and  the  ribosome  looks  for  the  next  tRNA  molecule.

Not  every  RNA  base  eventually  becomes  translated  into  a  protein,  and  so  an  interval  of  RNA  (or  an
interval  of  DNA  translated  into  RNA)  that  does  code  for  a  protein  is  of  great  biological  interest;;  such
an  interval  of  DNA  or  RNA  is  called  a  gene.  Because  protein  creation  drives  cellular  processes,
genes  differentiate  organisms  and  serve  as  a  basis  for  heredity,  or  the  process  by  which  traits  are
inherited.

Problem

The  20  commonly  occurring  amino  acids  are  abbreviated  by  using  20  letters  from  the  English  alphabet
(all  letters  except  for  B,  J,  O,  U,  X,  and  Z).  Protein  strings  are  constructed  from  these  20  symbols.
Henceforth,  the  term  genetic  string  will  incorporate  protein  strings  along  with  DNA  strings  and  RNA
strings.

The  RNA  codon  table  dictates  the  details  regarding  the  encoding  of  specific  codons  into  the  amino
 acid  alphabet.

Given:  An  RNA  string    corresponding  to  a  strand  of  mRNA  (of  length  at  most  10  kbp).
Return:  The  protein  string  encoded  by   .

Sample  Dataset

AUGGCCAUGGCGCCCAGAACUGAGAUCAAUAGUACCCGUAUUAACGGGUGA

Sample  Output

MAMAPRTEINSTRING

Problem  15
Finding  a  Motif  in  DNA

Combing  Through  the  Haystack

Finding  the  same  interval  of  DNA  in  the

genomes  of  two  different  organisms  (often
taken  from  different  species)  is  highly
suggestive  that  the  interval  has  the  same
function  in  both  organisms. Figure  1.  The  human  chromosomes
stained  with  a  probe  for  Alu
elements,  shown  in  green.
We  define  a  motif  as  such  a  commonly  shared  interval  of
DNA.  A  common  task  in  molecular  biology  is  to  search  an
organism's  genome  for  a  known  motif.

The  situation  is  complicated  by  the  fact  that  genomes  are  riddled  with  intervals  of  DNA  that  occur
multiple  times  (possibly  with  slight  modifications),  called  repeats.  These  repeats  occur  far  more
often  than  would  be  dictated  by  random  chance,  indicating  that  genomes  are  anything  but  random
and  in  fact  illustrate  that  the  language  of  DNA  must  be  very  powerful  (compare  with  the  frequent
reuse  of  common  words  in  any  human  language).

The  most  common  repeat  in  humans  is  the  Alu  repeat,  which  is  approximately  300  bp  long  and
recurs  around  a  million  times  throughout  every  human  genome  (see  Figure  1).  However,  Alu  has
not  been  found  to  serve  a  positive  purpose,  and  appears  in  fact  to  be  parasitic:  when  a  new  Alu
repeat  is  inserted  into  a  genome,  it  frequently  causes  genetic  disorders.

Problem

Given  two  strings    and   ,    is  a  substring  of    if    is  contained  as  a  contiguous  collection  of  symbols
in    (as  a  result,    must  be  no  longer  than   ).

The  position  of  a  symbol  in  a  string  is  the  total  number  of  symbols  found  to  its  left,  including  itself
(e.g.,  the  positions  of  all  occurrences  of  'U'  in  "AUGCUUCAGAAAGGUCUUACG"  are  2,  5,  6,  15,  17,
 and  18).  The  symbol  at  position    of    is  denoted  by   .

A  substring  of    can  be  represented  as   ,  where    and    represent  the  starting  and  ending
positions  of  the  substring  in   ;;  for  example,  if    =  "AUGCUUCAGAAAGGUCUUACG",  then    =
"UGCU".

The  location  of  a  substring    is  its  beginning  position   ;;  note  that    will  have  multiple  locations
in    if  it  occurs  more  than  once  as  a  substring  of    (see  the  Sample  below).

Given:  Two  DNA  strings    and    (each  of  length  at  most  1  kbp).
Return:  All  locations  of    as  a  substring  of   .

Sample  Dataset

GATATATGCATATACTT
ATAT

Sample  Output

2  4  10

Note

Different  programming  languages  use  different  notations  for  positions  of  symbols  in  strings.  Above,
we  use  1-­based  numbering,  as  opposed  to  0-­based  numbering,  which  is  used  in  Python.  For  
=  "AUGCUUCAGAAAGGUCUUACG",  1-­based  numbering  would  state  that    =  'A'  is  the  first
symbol  of  the  string,  whereas  this  symbol  is  represented  by    in  0-­based  numbering.  The  idea  of
0-­based  numbering  propagates  to  substring  indexing,  so  that    becomes  "GCUU"  instead  of
"UGCU".

Note  that  in  some  programming  languages,  such  as  Python,  s[j:k]  returns  only  fragment  from  index
 up  to  but  not  including  index   ,  so  that  s[2:5]  actually  becomes  "UGC",  not  "UGCU".

Problem  16
Consensus  and  Profile

Finding  a  Most  Likely  Common  Ancestor

In  “Counting  Point  Mutations”,  we  calculated  the  minimum  number  of  symbol
mismatches  between  two  strings  of  equal  length  to  model  the  problem  of  finding
the  minimum  number  of  point  mutations  occurring  on  the  evolutionary  path
between  two  homologous  strands  of  DNA.  If  we  instead  have  several  homologous
strands  that  we  wish  to  analyze  simultaneously,  then  the  natural  problem  is  to  find  an  average-­case
strand  to  represent  the  most  likely  common  ancestor  of  the  given  strands.
Problem

A  matrix  is  a  rectangular  table  of  values  divided  into  rows  and  columns.  An    matrix  has  
rows  and    columns.  Given  a  matrix   ,  we  write    to  indicate  the  value  found  at  the  intersection  of
row    and  column   .

Say  that  we  have  a  collection  of  DNA  strings,  all  having  the  same  length   .  Their  profile  matrix  is  a  
 matrix    in  which    represents  the  number  of  times  that  'A'  occurs  in  the   th  position  of
one  of  the  strings,    represents  the  number  of  times  that  C  occurs  in  the   th  position,  and  so  on
(see  below).

A  consensus  string    is  a  string  of  length    formed  from  our  collection  by  taking  the  most  common
symbol  at  each  position;;  the   th  symbol  of    therefore  corresponds  to  the  symbol  having  the  maximum
value  in  the   -­th  column  of  the  profile  matrix.  Of  course,  there  may  be  more  than  one  most  common
symbol,  leading  to  multiple  possible  consensus  strings.

A  T  C  C  A  G  C  T
G  G  G  C  A  A  C  T
A  T  G  G  A  T  C  T
DNA  Strings A  A  G  C  A  A  C  C
T  T  G  G  A  A  C  T
A  T  G  C  C  A  T  T
A  T  G  G  C  A  C  T

A      5  1  0  0  5  5  0  0
Profile C      0  0  1  4  2  0  6  1
G      1  1  6  3  0  1  0  0
T      1  5  0  0  0  1  1  6

Consensus A  T  G  C  A  A  C  T

Given:  A  collection  of  at  most  10  DNA  strings  of  equal  length  (at  most  1  kbp)  in  FASTA  format.
Return:  A  consensus  string  and  profile  matrix  for  the  collection.  (If  several  possible  consensus
strings  exist,  then  you  may  return  any  one  of  them.)

Sample  Dataset

>Rosalind_1
ATCCAGCT
>Rosalind_2
GGGCAACT
>Rosalind_3
ATGGATCT
>Rosalind_4
AAGCAACC
>Rosalind_5
TTGGAACT
>Rosalind_6
ATGCCATT
>Rosalind_7
 ATGGCACT

Sample  Output

ATGCAACT
A:  5  1  0  0  5  5  0  0
C:  0  0  1  4  2  0  6  1
G:  1  1  6  3  0  1  0  0
T:  1  5  0  0  0  1  1  6

Problem  17
Mortal  Fibonacci  Rabbits

Wabbit  Season

In  “Rabbits  and  Recurrence  Relations”,

we  mentioned  the  disaster  caused  by
introducing  European  rabbits  into
Australia.  By  the  turn  of  the  20th  Century,
the  situation  was  so  out  of  control  that  the  creatures  could
not  be  killed  fast  enough  to  slow  their  spread  (see  Figure
1).

The  conclusion?  Build  a  fence!  The  fence,  intended  to Figure  1.  A  c.1905  photo  from
preserve  the  sanctity  of  Western  Australia,  was  completed Australia  of  a  cart  loaded  to  the  hilt
with  rabbit  skins.
in  1907  after  undergoing  revisions  to  push  it  back  as  the
bunnies  pushed  their  frontier  ever  westward  (see  Figure  2).
If  it  sounds  like  a  crazy  plan,  the  Australians  at  the  time
seem  to  have  concurred,  as  shown  by  the  cartoon  in
Figure  3.

By  1950,  Australian  rabbits  numbered  600  million,  causing

the  government  to  decide  to  release  a  virus  (called
myxoma)  into  the  wild,  which  cut  down  the  rabbits  until
they  acquired  resistance.  In  a  final  Hollywood  twist,
another  experimental  rabbit  virus  escaped  in  1991,  and
some  resistance  has  already  been  observed. Figure  2.  Western  Australia's  rabbit
fence  is  actually  not  the  longest
The  bunnies  will  not  be  stopped,  but  they  don't  live  forever, fence  in  the  world  as  the  sign
claims.  That  honor  goes  to  a  3,500
and  so  in  this  problem,  our  aim  is  to  expand  Fibonacci's mile  fence  in  southeastern  Australia
rabbit  population  model  to  allow  for  mortal  rabbits. built  to  keep  out  dingoes.  Courtesy
Matt  Pounsett.

Problem

Recall  the  definition  of  the  Fibonacci  numbers  from  “Rabbits  and  Recurrence  Relations”,  which  followed
the  recurrence  relation    and  assumed  that  each  pair  of  rabbits  reaches  maturity
 in  one  month  and  produces  a  single  pair  of  offspring  (one
male,  one  female)  each  subsequent  month.

Our  aim  is  to  somehow  modify  this  recurrence  relation  to
achieve  a  dynamic  programming  solution  in  the  case  that  all
rabbits  die  out  after  a  fixed  number  of  months.  See  Figure  4
for  a  depiction  of  a  rabbit  tree  in  which  rabbits  live  for  three
months  (meaning  that  they  reproduce  only  twice  before  dying).

Given:  Positive  integers    and   . Figure  3.  An  1884  cartoon  from  the
Queensland  Figaro  proposing  how
Return:  The  total  number  of  pairs  of  rabbits  that  will  remain the  rabbits  viewed  their  fence.
after  the   -­th  month  if  all  rabbits  live  for    months.

Sample  Dataset

6  3

Sample  Output

4
Figure  4.  A  figure  illustrating  the
propagation  of  Fibonacci's  rabbits  if
they  die  after  three  months.

Problem  18
Overlap  Graphs

A  Brief  Introduction  to  Graph  Theory

Networks  arise  everywhere  in  the  practical  world,  especially  in  biology.  Networks
are  prevalent  in  popular  applications  such  as  modeling  the  spread  of  disease,  but
the  extent  of  network  applications  spreads  far  beyond  popular  science.  Our  first
question  asks  how  to  computationally  model  a  network  without  actually  needing
to  render  a  picture  of  the  network.

First,  some  terminology:  graph  is  the  technical  term  for  a  network;;  a  graph  is  made  up  of  hubs
called  nodes  (or  vertices),  pairs  of  which  are  connected  via  segments/curves  called  edges.  If  an
edge  connects  nodes    and   ,  then  it  is  denoted  by    (or  equivalently   ).

an  edge    is  incident  to  nodes    and   ;;  we  say  that    and    are  adjacent  to  each  other;;
the  degree  of    is  the  number  of  edges  incident  to  it;;
a  walk  is  an  ordered  collection  of  edges  for  which  the  ending  node  of  one  edge  is  the  starting
node  of  the  next  (e.g.,   ,   ,   ,  etc.);;
a  path  is  a  walk  in  which  every  node  appears  in  at  most  two  edges;;
path  length  is  the  number  of  edges  in  the  path;;
a  cycle  is  a  path  whose  final  node  is  equal  to  its  first  node  (so  that  every  node  is  incident  to
exactly  two  edges  in  the  cycle);;  and
 the  distance  between  two  vertices  is  the  length  of  the  shortest  path  connecting  them.

Graph  theory  is  the  abstract  mathematical  study  of  graphs  and  their  properties.

Problem

A  graph  whose  nodes  have  all  been  labeled  can  be  represented  by  an  adjacency  list,  in  which  each
row  of  the  list  contains  the  two  node  labels  corresponding  to  a  unique  edge.

A  directed  graph  (or  digraph)  is  a  graph  containing  directed  edges,  each  of  which  has  an
orientation.  That  is,  a  directed  edge  is  represented  by  an  arrow  instead  of  a  line  segment;;  the  starting
and  ending  nodes  of  an  edge  form  its  tail  and  head,  respectively.  The  directed  edge  with  tail    and
head    is  represented  by    (but  not  by   ).  A  directed  loop  is  a  directed  edge  of  the  form  
.

For  a  collection  of  strings  and  a  positive  integer   ,  the  overlap  graph  for  the  strings  is  a  directed
graph    in  which  each  string  is  represented  by  a  node,  and  string    is  connected  to  string    with  a
directed  edge  when  there  is  a  length    suffix  of    that  matches  a  length    prefix  of   ,  as  long  as  
;;  we  demand    to  prevent  directed  loops  in  the  overlap  graph  (although  directed  cycles  may
be  present).

Given:  A  collection  of  DNA  strings  in  FASTA  format  having  total  length  at  most  10  kbp.
Return:  The  adjacency  list  corresponding  to   .  You  may  return  edges  in  any  order.

Sample  Dataset

>Rosalind_0498
AAATAAA
>Rosalind_2391
AAATTTT
>Rosalind_2323
TTTTCCC
>Rosalind_0442
AAATCCC
>Rosalind_5013
GGGTGGG

Sample  Output

Rosalind_0498  Rosalind_2391
Rosalind_0498  Rosalind_0442
Rosalind_2391  Rosalind_2323

Note  on  Visualizing  Graphs

If  you  are  looking  for  a  way  to  actually  visualize  graphs  as  you  are  working  through  the  Rosalind
site,  then  you  may  like  to  consider  Graphviz  (link  here),  a  cross-­platform  application  for  rendering
graphs.
Problem  19
Calculating  Expected  Offspring

The  Need  for  Averages

Averages  arise  everywhere.  In  sports,  we  want  to  project  the  average  number  of
games  that  a  team  is  expected  to  win;;  in  gambling,  we  want  to  project  the
average  losses  incurred  playing  blackjack;;  in  business,  companies  want  to
calculate  their  average  expected  sales  for  the  next  quarter.

Molecular  biology  is  not  immune  from  the  need  for  averages.  Researchers  need  to  predict  the
expected  number  of  antibiotic-­resistant  pathogenic  bacteria  in  a  future  outbreak,  estimate  the
predicted  number  of  locations  in  the  genome  that  will  match  a  given  motif,  and  study  the
distribution  of  alleles  throughout  an  evolving  population.  In  this  problem,  we  will  begin  discussing
the  third  issue;;  first,  we  need  to  have  a  better  understanding  of  what  it  means  to  average  a  random
process.

Problem

For  a  random  variable    taking  integer  values  between  1  and   ,  the  expected  value  of    is  
.  The  expected  value  offers  us  a  way  of  taking  the  long-­term  average
of  a  random  variable  over  a  large  number  of  trials.

As  a  motivating  example,  let    be  the  number  on  a  six-­sided  die.  Over  a  large  number  of  rolls,  we
should  expect  to  obtain  an  average  of  3.5  on  the  die  (even  though  it's  not  possible  to  roll  a  3.5).  The
formula  for  expected  value  confirms  that   .

More  generally,  a  random  variable  for  which  every  one  of  a  number  of  equally  spaced  outcomes  has  the
same  probability  is  called  a  uniform  random  variable  (in  the  die  example,  this  "equal  spacing"  is
equal  to  1).  We  can  generalize  our  die  example  to  find  that  if    is  a  uniform  random  variable  with
minimum  possible  value    and  maximum  possible  value   ,  then   .  You  may  also  wish  to
verify  that  for  the  dice  example,  if    is  the  random  variable  associated  with  the  outcome  of  a  second
die  roll,  then   .

Given:  Six  positive  integers,  each  of  which  does  not  exceed  20,000.  The  integers  correspond  to  the
number  of  couples  in  a  population  possessing  each  genotype  pairing  for  a  given  factor.  In  order,  the  six
given  integers  represent  the  number  of  couples  having  the  following  genotypes:

1.   AA-­AA
2.   AA-­Aa
3.   AA-­aa
4.   Aa-­Aa
5.   Aa-­aa
6.   aa-­aa

Return:  The  expected  number  of  offspring  displaying  the  dominant  phenotype  in  the  next
generation,  under  the  assumption  that  every  couple  has  exactly  two  offspring.

Sample  Dataset
 1  0  0  1  0  1

Sample  Output

3.5

Problem  20
Finding  a  Shared  Motif

Searching  Through  the  Haystack

In  “Finding  a  Motif  in  DNA”,  we  searched  a  given  genetic  string  for  a  motif;;
however,  this  problem  assumed  that  we  know  the  motif  in  advance.  In  practice,
biologists  often  do  not  know  exactly  what  they  are  looking  for.  Rather,  they  must
hunt  through  several  different  genomes  at  the  same  time  to  identify  regions  of
similarity  that  may  indicate  genes  shared  by  different  organisms  or  species.

The  simplest  such  region  of  similarity  is  a  motif  occurring  without  mutation  in  every  one  of  a
collection  of  genetic  strings  taken  from  a  database;;  such  a  motif  corresponds  to  a  substring  shared
by  all  the  strings.  We  want  to  search  for  long  shared  substrings,  as  a  longer  motif  will  likely
indicate  a  greater  shared  function.

Problem

A  common  substring  of  a  collection  of  strings  is  a  substring  of  every  member  of  the  collection.  We
say  that  a  common  substring  is  a  longest  common  substring  if  there  does  not  exist  a  longer
common  substring.  For  example,  "CG"  is  a  common  substring  of  "ACGTACGT"  and  "AACCGGTATA",
but  it  is  not  as  long  as  possible;;  in  this  case,  "GTA"  is  a  longest  common  substring  of  "ACGTACGT"
and  "AACCGTATA".

Note  that  the  longest  common  substring  is  not  necessarily  unique;;  for  a  simple  example,  "AA"  and
"CC"  are  both  longest  common  substrings  of  "AACC"  and  "CCAA".

Given:  A  collection  of    ( )  DNA  strings  of  length  at  most  1  kbp  each  in  FASTA  format.

Return:  A  longest  common  substring  of  the  collection.  (If  multiple  solutions  exist,  you  may  return
any  single  solution.)

Sample  Dataset

>Rosalind_1
GATTACA
>Rosalind_2
TAGACCA
>Rosalind_3
 ATACA

Sample  Output

AC

Problem  21
Independent  Alleles

Mendel's  Second  Law

Recall  that  Mendel's  first  law  states  that

for  any  factor,  an  individual  randomly
assigns  one  of  its  two  alleles  to  its
offspring.  Yet  this  law  does  not  state
anything  regarding  the  relationship  with  which  alleles  for
different  factors  will  be  inherited.

After  recording  the  results  of  crossing  thousands  of  pea

plants  for  seven  years,  Mendel  surmised  that  alleles  for
different  factors  are  inherited  with  no  dependence  on  each
other.  This  statement  has  become  his  second  law,  also
known  as  the  law  of  independent  assortment.

What  does  it  mean  for  factors  to  be  "assorted Figure  1.  Mendel's  second  law
dictates  that  every  one  of  the  16
independently?"  If  we  cross  two  organisms,  then  a possible  assignments  of  parental
shortened  form  of  independent  assortment  states  that  if  we alleles  is  equally  likely.  The  Punnett
look  only  at  organisms  having  the  same  alleles  for  one square  for  two  factors  therefore
places  each  of  these  assignments
factor,  then  the  inheritance  of  another  factor  should  not in  a  cell  of  a  4  X  4  table.  The
change. probability  of  an  offspring's  genome
is  equal  to  the  number  of  times  it
For  example,  Mendel's  first  law  states  that  if  we  cross  two   appears  in  the  table,  divided  by  16.
 organisms,  then  1/4  of  their  offspring  will  be   ,  1/4
will  be   ,  and  1/2  will  be   .  Now,  say  that  we  cross  plants  that  are  both  heterozygous  for  two
factors,  so  that  both  of  their  genotypes  may  be  written  as   .  Next,  examine  only  
offspring:  Mendel's  second  law  states  that  the  same  proportions  of   ,   ,  and    individuals  will
be  observed  in  these  offspring.  The  same  fact  holds  for    and    offspring.

As  a  result,  independence  will  allow  us  to  say  that  the  probability  of  an    offspring  is  simply
equal  to  the  probability  of  an    offspring  times  the  probability  of  a    organism,  i.e.,  1/16.

Because  of  independence,  we  can  also  extend  the  idea  of  Punnett  squares  to  multiple  factors,  as
shown  in  Figure  1.  We  now  wish  to  quantify  Mendel's  notion  of  independence  using  probability.

Problem

Two  events    and    are  independent  if    is  equal  to   .  In  other  words,
the  events  do  not  influence  each  other,  so  that  we  may  simply
calculate  each  of  the  individual  probabilities  separately  and
then  multiply.

More  generally,  random  variables    and    are  independent

if  whenever    and    are  respective  events  for    and   ,  
and    are  independent  (i.e.,  
).

As  an  example  of  how  helpful  independence  can  be  for

calculating  probabilities,  let    and    represent  the  numbers Figure  2.  The  probability  of  each
showing  on  two  six-­sided  dice.  Intuitively,  the  number  of  pips outcome  for  the  sum  of  the  values
on  two  rolled  dice  (black  and  white),
showing  on  one  die  should  not  affect  the  number  showing  on broken  down  depending  on  the
the  other  die.  If  we  want  to  find  the  probability  that    is number  of  pips  showing  on  each
die.  You  can  verify  that  18  of  the  36
odd,  then  we  don't  need  to  draw  a  tree  diagram  and  consider
equally  probable  possibilities  result
all  possibilities.  We  simply  first  note  that  for    to  be in  an  odd  sum.
odd,  either    is  even  and    is  odd  or    is  odd  and    is
even.  In  terms  of  probability,  
.  Using
independence,  this  becomes  

,  or   .
You  can  verify  this  result  in  Figure  2,  which  shows  all  36  outcomes  for  rolling  two  dice.

Given:  Two  positive  integers    ( )  and    ( ).  In  this  problem,  we  begin  with  Tom,  who
in  the  0th  generation  has  genotype  Aa  Bb.  Tom  has  two  children  in  the  1st  generation,  each  of  whom
has  two  children,  and  so  on.  Each  organism  always  mates  with  an  organism  having  genotype  Aa  Bb.

Return:  The  probability  that  at  least    Aa  Bb  organisms  will  belong  to  the   -­th  generation  of  Tom's
family  tree  (don't  count  the  Aa  Bb  mates  at  each  level).  Assume  that  Mendel's  second  law  holds  for
the  factors.

Sample  Dataset

2  1

Sample  Output

0.684

An  Example  of  Dependent  Random  Variables

Two  random  variables  are  dependent  if  they  are  not

independent.  For  an  example  of  dependent  random
variables,  recall  our  example  in  “Mendel's  First  Law”  of
drawing  two  balls  from  a  bag  containing  3  red  balls  and  2
blue  balls.  If    represents  the  color  of  the  first  ball  drawn
and    is  the  color  of  the  second  ball  drawn  (without
replacement),  then  let    be  the  event  "  is  red"  and    be
the  event    is  blue.  In  this  case,  the  probability  tree
Figure  3.  The  probability  of  any
diagram  illustrated  in  Figure  3  demonstrates  that   outcome  (leaf)  in  a  probability  tree
.  Yet  we  can  also  see  that   diagram  is  given  by  the  product  of
probabilities  from  the  start  of  the  tree
  and   .  We  can  now to  the  outcome.  For  example,  the
probability  that  X  is  blue  and  Y  is  red
see  that   . is  equal  to  (2/5)(1/4),  or  1/10.

Problem  22
Finding  a  Protein  Motif

Motif  Implies  Function

As  mentioned  in  “Translating  RNA  into

Protein”,  proteins  perform  every  practical
function  in  the  cell.  A  structural  and
functional  unit  of  the  protein  is  a  domain:
in  terms  of  the  protein's  primary  structure,  the  domain  is  an
interval  of  amino  acids  that  can  evolve  and  function
independently.

Each  domain  usually  corresponds  to  a  single  function  of

the  protein  (e.g.,  binding  the  protein  to  DNA,  creating  or
breaking  specific  chemical  bonds,  etc.).  Some  proteins,
such  as  myoglobin  and  the  Cytochrome  complex,  have
only  one  domain,  but  many  proteins  are  multifunctional  and Figure  1.  The  human  cyclophilin
therefore  possess  several  domains.  It  is  even  possible  to family,  as  represented  by  the
artificially  fuse  different  domains  into  a  protein  molecule structures  of  the  isomerase
domains  of  some  of  its  members.
with  definite  properties,  creating  a  chimeric  protein.

Just  like  species,  proteins  can  evolve,  forming  homologous  groups  called  protein  families.
Proteins  from  one  family  usually  have  the  same  set  of  domains,  performing  similar  functions;;  see
Figure  1.

A  component  of  a  domain  essential  for  its  function  is  called  a  motif,  a  term  that  in  general  has  the
same  meaning  as  it  does  in  nucleic  acids,  although  many  other  terms  are  also  used  (blocks,
signatures,  fingerprints,  etc.)  Usually  protein  motifs  are  evolutionarily  conservative,  meaning  that
they  appear  without  much  change  in  different  species.

Proteins  are  identified  in  different  labs  around  the  world  and  gathered  into  freely  accessible
databases.  A  central  repository  for  protein  data  is  UniProt,  which  provides  detailed  protein
annotation,  including  function  description,  domain  structure,  and  post-­translational  modifications.
UniProt  also  supports  protein  similarity  search,  taxonomy  analysis,  and  literature  citations.

Problem

To  allow  for  the  presence  of  its  varying  forms,  a  protein  motif  is  represented  by  a  shorthand  as  follows:
[XY]  means  "either  X  or  Y"  and  {X}  means  "any  amino  acid  except  X."  For  example,  the  N-­
glycosylation  motif  is  written  as  N{P}[ST]{P}.

You  can  see  the  complete  description  and  features  of  a  particular  protein  by  its  access  ID  "uniprot_id"
in  the  UniProt  database,  by  inserting  the  ID  number  into
 http://www.uniprot.org/uniprot/uniprot_id

Alternatively,  you  can  obtain  a  protein  sequence  in  FASTA  format  by  following

http://www.uniprot.org/uniprot/uniprot_id.fasta

For  example,  the  data  for  protein  B5ZC00  can  be  found  at  http://www.uniprot.org/uniprot/B5ZC00.

Given:  At  most  15  UniProt  Protein  Database  access  IDs.

Return:  For  each  protein  possessing  the  N-­glycosylation  motif,  output  its  given  access  ID  followed
by  a  list  of  locations  in  the  protein  string  where  the  motif  can  be  found.

Sample  Dataset

A2Z669
B5ZC00
P07204_TRBM_HUMAN
P20840_SAG1_YEAST

Sample  Output

B5ZC00
85  118  142  306  395
P07204_TRBM_HUMAN
47  115  116  382  409
P20840_SAG1_YEAST
79  109  135  248  306  348  364  402  485  501  614

Note

Some  entries  in  UniProt  have  one  primary  (citable)  accession  number  and  some  secondary
numbers,  appearing  due  to  merging  or  demerging  entries.  In  this  problem,  you  may  be  given  any
type  of  ID.  If  you  type  the  secondary  ID  into  the  UniProt  query,  then  you  will  be  automatically
redirected  to  the  page  containing  the  primary  ID.  You  can  find  more  information  about  UniProt  IDs
here.

Problem  23
Inferring  mRNA  from  Protein

Pitfalls  of  Reversing  Translation

When  researchers  discover  a  new  protein,  they  would  like  to  infer  the  strand  of  mRNA  from  which
this  protein  could  have  been  translated,  thus  allowing  them  to  locate  genes  associated  with  this
 protein  on  the  genome.

Unfortunately,  although  any  RNA  string  can  be  translated  into  a  unique  protein
string,  reversing  the  process  yields  a  huge  number  of  possible  RNA  strings  from  a
single  protein  string  because  most  amino  acids  correspond  to  multiple  RNA
codons  (see  the  RNA  Codon  Table).

Because  of  memory  considerations,  most  data  formats  that  are  built  into
languages  have  upper  bounds  on  how  large  an  integer  can  be:  in  some  versions  of  Python,  an  "int"
variable  may  be  required  to  be  no  larger  than   ,  or  2,147,483,647.  As  a  result,  to  deal  with
very  large  numbers  in  Rosalind,  we  need  to  devise  a  system  that  allows  us  to  manipulate  large
numbers  without  actually  having  to  store  large  numbers.

Problem

For  positive  integers    and   ,    modulo    (written    in  shorthand)  is  the  remainder  when  
 is  divided  by   .  For  example,    because   .

Modular  arithmetic  is  the  study  of  addition,  subtraction,  multiplication,  and  division  with  respect  to
the  modulo  operation.  We  say  that    and    are  congruent  modulo    if   ;;
in  this  case,  we  use  the  notation   .

Two  useful  facts  in  modular  arithmetic  are  that  if    and   ,  then  
 and   .  To  check  your  understanding  of  these
rules,  you  may  wish  to  verify  these  relationships  for   ,   ,   ,   ,  and   .

As  you  will  see  in  this  exercise,  some  Rosalind  problems  will  ask  for  a  (very  large)  integer  solution
modulo  a  smaller  number  to  avoid  the  computational  pitfalls  that  arise  with  storing  such  large  numbers.

Given:  A  protein  string  of  length  at  most  1000  aa.

Return:  The  total  number  of  different  RNA  strings  from  which  the  protein  could  have  been
translated,  modulo  1,000,000.  (Don't  neglect  the  importance  of  the  stop  codon  in  protein  translation.)

Sample  Dataset

MA

Sample  Output

12

Hint

What  does  it  mean  intuitively  to  take  a  number  modulo  1,000,000?

Problem  24
Open  Reading  Frames
 Transcription  May  Begin  Anywhere

In  “Transcribing  DNA  into  RNA”,  we

discussed  the  transcription  of  DNA  into
RNA,  and  in  “Translating  RNA  into
Protein”,  we  examined  the  translation  of
RNA  into  a  chain  of  amino  acids  for  the  construction  of
proteins.  We  can  view  these  two  processes  as  a  single
step  in  which  we  directly  translate  a  DNA  string  into  a
protein  string,  thus  calling  for  a  DNA  codon  table.

However,  three  immediate  wrinkles  of  complexity  arise

Figure  1.  Schematic  image  of  the
when  we  try  to  pass  directly  from  DNA  to  proteins.  First, particular  ORF  with  start  and  stop
not  all  DNA  will  be  transcribed  into  RNA:  so-­called  junk codons  shown.
DNA  appears  to  have  no  practical  purpose  for  cellular
function.  Second,  we  can  begin  translation  at  any  position  along  a  strand  of  RNA,  meaning  that  any
substring  of  a  DNA  string  can  serve  as  a  template  for  translation,  as  long  as  it  begins  with  a  start
codon,  ends  with  a  stop  codon,  and  has  no  other  stop  codons  in  the  middle.  See  Figure  1.  As  a
result,  the  same  RNA  string  can  actually  be  translated  in  three  different  ways,  depending  on  how
we  group  triplets  of  symbols  into  codons.  For  example,  ...AUGCUGAC...  can  be  translated  as
...AUGCUG...,  ...UGCUGA...,  and  ...GCUGAC...,  which  will  typically  produce  wildly  different
protein  strings.

Problem

Either  strand  of  a  DNA  double  helix  can  serve  as  the  coding  strand  for  RNA  transcription.  Hence,  a
given  DNA  string  implies  six  total  reading  frames,  or  ways  in  which  the  same  region  of  DNA  can  be
translated  into  amino  acids:  three  reading  frames  result  from  reading  the  string  itself,  whereas  three
more  result  from  reading  its  reverse  complement.

An  open  reading  frame  (ORF)  is  one  which  starts  from  the  start  codon  and  ends  by  stop  codon,
without  any  other  stop  codons  in  between.  Thus,  a  candidate  protein  string  is  derived  by  translating  an
open  reading  frame  into  amino  acids  until  a  stop  codon  is  reached.

Given:  A  DNA  string    of  length  at  most  1  kbp  in  FASTA  format.
Return:  Every  distinct  candidate  protein  string  that  can  be  translated  from  ORFs  of   .  Strings  can
be  returned  in  any  order.

Sample  Dataset

>Rosalind_99
AGCCATGTAGCTAACTCAGGTTACATGGGGATGACCCCGCGACTTGGATTAGAGTCTCTTTTGGAATAAGC
CTGAATGATCCGAGTAGCATCTCAG

Sample  Output

MLLGSFRLIPKETLIQVAGSSPCNLS
M
MGMTPRLGLESLLE
 MTPRLGLESLLE

Problem  25
Enumerating  Gene  Orders

Rearrangements  Power  Large-­Scale  Genomic  Changes

Point  mutations  can  create  changes  in

populations  of  organisms  from  the  same
species,  but  they  lack  the  power  to
create  and  differentiate  entire  species.
This  more  arduous  work  is  left  to  larger  mutations  called
genome  rearrangements,  which  move  around  huge
blocks  of  DNA.  Rearrangements  cause  major  genomic
change,  and  most  rearrangements  are  fatal  or  seriously
damaging  to  the  mutated  cell  and  its  descendants  (many Figure  1.  Similar  regions  in  mouse
cancers  derive  from  rearrangements).  For  this  reason, and  human  chromosomes.  Image
credit:  U.S.  Department  of  Energy
rearrangements  that  come  to  influence  the  genome  of  an Human  Genome  Program
entire  species  are  very  rare.

Because  rearrangements  that  affect  species  evolution  occur  infrequently,  two  closely  related
species  will  have  very  similar  genomes.  Thus,  to  simplify  comparison  of  two  such  genomes,
researchers  first  identify  similar  intervals  of  DNA  from  the  species,  called  synteny  blocks;;  over
time,  rearrangements  have  created  these  synteny  blocks  and  heaved  them  around  across  the  two
genomes  (often  separating  blocks  onto  different  chromosomes,  see  Figure  1.).

A  pair  of  synteny  blocks  from  two  different  species  are  not  strictly  identical  (they  are  separated  by
the  action  of  point  mutations  or  very  small  rearrangements),  but  for  the  sake  of  studying  large-­scale
rearrangements,  we  consider  them  to  be  equivalent.  As  a  result,  we  can  label  each  synteny  block
with  a  positive  integer;;  when  comparing  two  species'  genomes/chromosomes,  we  then  only  need  to
specify  the  order  of  its  numbered  synteny  blocks.

Problem

A  permutation  of  length    is  an  ordering  of  the  positive  integers   .  For  example,  
 is  a  permutation  of  length   .

Given:  A  positive  integer   .

Return:  The  total  number  of  permutations  of  length   ,  followed  by  a  list  of  all  such  permutations  (in
any  order).

Sample  Dataset

3
 Sample  Output

6
1  2  3
1  3  2
2  1  3
2  3  1
3  1  2
3  2  1

Problem  26
Calculating  Protein  Mass

Chaining  the  Amino  Acids

In  “Translating  RNA  into  Protein”,  we

examined  the  translation  of  RNA  into  an
amino  acid  chain  for  the  construction  of  a
protein.  When  two  amino  acids  link
together,  they  form  a  peptide  bond,  which  releases  a
molecule  of  water;;  see  Figure  1.  Thus,  after  a  series  of
amino  acids  have  been  linked  together  into  a  polypeptide,
every  pair  of  adjacent  amino  acids  has  lost  one  molecule  of
water,  meaning  that  a  polypeptide  containing    amino
acids  has  had    water  molecules  removed.
Figure  1.  Formation  of  a  peptide
More  generally,  a  residue  is  a  molecule  from  which  a bond
water  molecule  has  been  removed;;  every  amino  acid  in  a
protein  are  residues  except  the  leftmost  and  the  rightmost
ones.  These  outermost  amino  acids  are  special  in  that  one
has  an  "unstarted"  peptide  bond,  and  the  other  has  an
"unfinished"  peptide  bond.  Between  them,  the  two
molecules  have  a  single  "extra"  molecule  of  water  (see  the
atoms  marked  in  blue  in  Figure  2).  Thus,  the  mass  of  a
protein  is  the  sum  of  masses  of  all  its  residues  plus  the
mass  of  a  single  water  molecule.

There  are  two  standard  ways  of  computing  the  mass  of  a
residue  by  summing  the  masses  of  its  individual  atoms.  Its Figure  2.  Outermost  acids
monoisotopic  mass  is  computed  by  using  the  principal
(most  abundant)  isotope  of  each  atom  in  the  amino  acid,  whereas  its  average  mass  is  taken  by
taking  the  average  mass  of  each  atom  in  the  molecule  (over  all  naturally  appearing  isotopes).

Many  applications  in  proteomics  rely  on  mass  spectrometry,  an  analytical  chemical  technique
used  to  determine  the  mass,  elemental  composition,  and  structure  of  molecules.  In  mass
spectrometry,  monoisotopic  mass  is  used  more  often  than  average  mass,  and  so  all  amino  acid
masses  are  assumed  to  be  monoisotopic  unless  otherwise  stated.
 The  standard  unit  used  in  mass  spectrometry  for  measuring  mass  is  the  atomic  mass  unit,  which
is  also  called  the  dalton  (Da)  and  is  defined  as  one  twelfth  of  the  mass  of  a  neutral  atom  of  carbon-­
12.  The  mass  of  a  protein  is  the  sum  of  the  monoisotopic  masses  of  its  amino  acid  residues  plus
the  mass  of  a  single  water  molecule  (whose  monoisotopic  mass  is  18.01056  Da).

In  the  following  several  problems  on  applications  of  mass  spectrometry,  we  avoid  the  complication
of  having  to  distinguish  between  residues  and  non-­residues  by  only  considering  peptides  excised
from  the  middle  of  the  protein.  This  is  a  relatively  safe  assumption  because  in  practice,  peptide
analysis  is  often  performed  in  tandem  mass  spectrometry.  In  this  special  class  of  mass
spectrometry,  a  protein  is  first  divided  into  peptides,  which  are  then  broken  into  ions  for  mass
analysis.

Problem

In  a  weighted  alphabet,  every  symbol  is  assigned  a  positive  real  number  called  a  weight.  A  string
formed  from  a  weighted  alphabet  is  called  a  weighted  string,  and  its  weight  is  equal  to  the  sum  of
the  weights  of  its  symbols.

The  standard  weight  assigned  to  each  member  of  the  20-­symbol  amino  acid  alphabet  is  the
monoisotopic  mass  of  the  corresponding  amino  acid.

Given:  A  protein  string    of  length  at  most  1000  aa.

Return:  The  total  weight  of   .  Consult  the  monoisotopic  mass  table.

Sample  Dataset

SKADYEK

Sample  Output

821.392

Problem  27
Locating  Restriction  Sites

The  Billion-­Year  War

The  war  between  viruses  and  bacteria  has  been  waged  for  over  a  billion  years.
Viruses  called  bacteriophages  (or  simply  phages)  require  a  bacterial  host  to
propagate,  and  so  they  must  somehow  infiltrate  the  bacterium;;  such  deception
can  only  be  achieved  if  the  phage  understands  the  genetic  framework  underlying
the  bacterium's  cellular  functions.  The  phage's  goal  is  to  insert  DNA  that  will  be  replicated  within
the  bacterium  and  lead  to  the  reproduction  of  as  many  copies  of  the  phage  as  possible,  which
sometimes  also  involves  the  bacterium's  demise.
 To  defend  itself,  the  bacterium  must  either  obfuscate  its
cellular  functions  so  that  the  phage  cannot  infiltrate  it,  or
better  yet,  go  on  the  counterattack  by  calling  in  the  air
force.  Specifically,  the  bacterium  employs  aerial  scouts
called  restriction  enzymes,  which  operate  by  cutting
through  viral  DNA  to  cripple  the  phage.  But  what  kind  of
DNA  are  restriction  enzymes  looking  for?

The  restriction  enzyme  is  a  homodimer,  which  means

that  it  is  composed  of  two  identical  substructures.  Each  of
these  structures  separates  from  the  restriction  enzyme  in
order  to  bind  to  and  cut  one  strand  of  the  phage  DNA Figure  1.  DNA  cleaved  by  EcoRV
molecule;;  both  substructures  are  pre-­programmed  with  the restriction  enzyme
same  target  string  containing  4  to  12  nucleotides  to  search
for  within  the  phage  DNA  (see  Figure  1.).  The  chance  that  both  strands  of  phage  DNA  will  be  cut
(thus  crippling  the  phage)  is  greater  if  the  target  is  located  on  both  strands  of  phage  DNA,  as  close
to  each  other  as  possible.  By  extension,  the  best  chance  of  disarming  the  phage  occurs  when  the
two  target  copies  appear  directly  across  from  each  other  along  the  phage  DNA,  a  phenomenon  that
occurs  precisely  when  the  target  is  equal  to  its  own  reverse  complement.  Eons  of  evolution  have
made  sure  that  most  restriction  enzyme  targets  now  have  this  form.

Problem

A  DNA  string  is  a  reverse  palindrome  if  it  is  equal  to  its
reverse  complement.  For  instance,  GCATGC  is  a  reverse
palindrome  because  its  reverse  complement  is  GCATGC.  See
Figure  2.
Figure  2.  Palindromic  recognition
Given:  A  DNA  string  of  length  at  most  1  kbp  in  FASTA site
format.

Return:  The  position  and  length  of  every  reverse  palindrome  in  the  string  having  length  between  4
and  12.  You  may  return  these  pairs  in  any  order.

Sample  Dataset

>Rosalind_24
TCAATGCATGCGGGTCTATATGCAT

Sample  Output

4  6
5  4
6  6
7  4
17  4
18  4
20  6
21  4
 Extra  Information

You  may  be  curious  how  the  bacterium  prevents  its  own  DNA  from  being  cut  by  restriction
enzymes.  The  short  answer  is  that  it  locks  itself  from  being  cut  through  a  chemical  process  called
DNA  methylation.

Problem  28
RNA  Splicing

Genes  are  Discontiguous

In  “Transcribing  DNA  into  RNA”,  we

mentioned  that  a  strand  of  DNA  is  copied
into  a  strand  of  RNA  during  transcription,
Figure  1.  The  elongation  of  a  pre-­
but  we  neglected  to  mention  how
mRNA  by  RNAP  as  it  moves  down
transcription  is  achieved. the  template  strand  of  DNA.

In  the  nucleus,  an  enzyme  (i.e.,  a  molecule  that

accelerates  a  chemical  reaction)  called  RNA  polymerase
(RNAP)  initiates  transcription  by  breaking  the  bonds  joining
complementary  bases  of  DNA.  It  then  creates  a  molecule
called  precursor  mRNA,  or  pre-­mRNA,  by  using  one  of
Figure  2.  RNA  is  identical  to  the
the  two  strands  of  DNA  as  a  template  strand:  moving coding  strand  except  for  the
down  the  template  strand,  when  RNAP  encounters  the replacement  of  thymine  with  uracil.
next  nucleotide,  it  adds  the  complementary  base  to  the
growing  RNA  strand,  with  the  provision  that  uracil  must  be  used  in  place  of  thymine;;  see  Figure  1.

Because  RNA  is  constructed  based  on  complementarity,  the  second  strand  of  DNA,  called  the
coding  strand,  is  identical  to  the  new  strand  of  RNA  except  for  the  replacement  of  thymine  with
uracil.  See  Figure  2  and  recall  “Transcribing  DNA  into  RNA”.

After  RNAP  has  created  several  nucleotides  of  RNA,  the  first  separated  complementary  DNA  bases
then  bond  back  together.  The  overall  effect  is  very  similar  to  a  pair  of  zippers  traversing  the  DNA
double  helix,  unzipping  the  two  strands  and  then  quickly  zipping  them  back  together  while  the
strand  of  pre-­mRNA  is  produced.

For  that  matter,  it  is  not  the  case  that  an  entire  substring  of  DNA  is  transcribed  into  RNA  and  then
translated  into  a  peptide  one  codon  at  a  time.  In  reality,  a  pre-­mRNA  is  first  chopped  into  smaller
segments  called  introns  and  exons;;  for  the  purposes  of  protein  translation,  the  introns  are  thrown
out,  and  the  exons  are  glued  together  sequentially  to  produce  a  final  strand  of  mRNA.  This  cutting
and  pasting  process  is  called  splicing,  and  it  is  facilitated  by  a  collection  of  RNA  and  proteins
called  a  spliceosome.  The  fact  that  the  spliceosome  is  made  of  RNA  and  proteins  despite
regulating  the  splicing  of  RNA  to  create  proteins  is  just  one  manifestation  of  a  molecular  chicken-­
and-­egg  scenario  that  has  yet  to  be  fully  resolved.

In  terms  of  DNA,  the  exons  deriving  from  a  gene  are  collectively  known  as  the  gene's  coding
region.
 Problem

After  identifying  the  exons  and  introns  of  an  RNA  string,  we  only  need  to  delete  the  introns  and
concatenate  the  exons  to  form  a  new  string  ready  for  translation.

Given:  A  DNA  string    (of  length  at  most  1  kbp)  and  a  collection  of  substrings  of    acting  as
introns.  All  strings  are  given  in  FASTA  format.

Return:  A  protein  string  resulting  from  transcribing  and  translating  the  exons  of   .  (Note:  Only  one
solution  will  exist  for  the  dataset  provided.)

Sample  Dataset

>Rosalind_10
ATGGTCTACATAGCTGACAAACAGCACGTAGCAATCGGTCGAATCTCGAGAGGCATATGGTCACATGATCG
GTCGAGCGTGTTTCAAAGTTTGCGCCTAG
>Rosalind_12
ATCGGTCGAA
>Rosalind_15
ATCGGTCGAGCGTGT

Sample  Output

MVYIADKQHVASREAYGHMFKVCA

Problem  29
Enumerating  k-­mers  Lexicographically

Organizing  Strings

When  cataloguing  a  collection  of  genetic  strings,  we  should  have  an  established
system  by  which  to  organize  them.  The  standard  method  is  to  organize  strings  as
they  would  appear  in  a  dictionary,  so  that  "APPLE"  precedes  "APRON",  which  in
turn  comes  before  "ARMOR".

Problem

Assume  that  an  alphabet    has  a  predetermined  order;;  that  is,  we  write  the  alphabet  as  a
permutation   ,  where   .  For  instance,  the  English
alphabet  is  organized  as   .

Given  two  strings    and    having  the  same  length   ,  we  say  that    precedes    in  the  lexicographic
order  (and  write   )  if  the  first  symbol    that  doesn't  match    satisfies    in   .

Given:  A  collection  of  at  most  10  symbols  defining  an  ordered  alphabet,  and  a  positive  integer    (
 ).

Return:  All  strings  of  length    that  can  be  formed  from  the  alphabet,  ordered  lexicographically.

Sample  Dataset

T  A  G  C
2

Sample  Output

TT
TA
TG
TC
AT
AA
AG
AC
GT
GA
GG
GC
CT
CA
CG
CC

Note

As  illustrated  in  the  sample,  the  alphabet  order  in  this  problem  is  defined  by  the  order  in  which
symbols  are  provided  in  the  dataset,  which  is  not  necessarily  the  traditional  order  of  the  English
alphabet.

Problem  30
Longest  Increasing  Subsequence

A  Simple  Measure  of  Gene  Order  Similarity

In  “Enumerating  Gene  Orders”,  we  started  talking  about  comparing  the  order  of
genes  on  a  chromosome  taken  from  two  different  species  and  moved  around  by
rearrangements  throughout  the  course  of  evolution.

One  very  simple  way  of  comparing  genes  from  two  chromosomes  is  to  search  for  the  largest
 collection  of  genes  that  are  found  in  the  same  order  in  both  chromosomes.  To  do  so,  we  will  need
to  apply  our  idea  of  permutations.  Say  that  two  chromosomes  share    genes;;  if  we  label  the  genes
of  one  chromosome  by  the  numbers  1  through  n  in  the  order  that  they  appear,  then  the  second
chromosome  will  be  given  by  a  permutation  of  these  numbered  genes.  To  find  the  largest  number  of
genes  appearing  in  the  same  order,  we  need  only  to  find  the  largest  collection  of  increasing
elements  in  the  permutation.

Problem

A  subsequence  of  a  permutation  is  a  collection  of  elements  of  the  permutation  in  the  order  that  they
appear.  For  example,  (5,  3,  4)  is  a  subsequence  of  (5,  1,  3,  4,  2).

A  subsequence  is  increasing  if  the  elements  of  the  subsequence  increase,  and  decreasing  if  the
elements  decrease.  For  example,  given  the  permutation  (8,  2,  1,  6,  5,  7,  4,  3,  9),  an  increasing
subsequence  is  (2,  6,  7,  9),  and  a  decreasing  subsequence  is  (8,  6,  5,  4,  3).  You  may  verify  that  these
two  subsequences  are  as  long  as  possible.

Given:  A  positive  integer    followed  by  a  permutation    of  length   .

Return:  A  longest  increasing  subsequence  of   ,  followed  by  a  longest  decreasing  subsequence  of  
.

Sample  Dataset

5
5  1  4  2  3

Sample  Output

1  2  3
5  4  2

Citation

Adapted  from  Jones  &  Pevzner,  *An  Introduction  to  Bioinformatics  Algorithms,  Problem  6.48.

Problem  31
Genome  Assembly  as  Shortest  Superstring

Introduction  to  Genome  Sequencing

Recall  from  “Computing  GC  Content”  that  almost  all  humans  share  approximately
99.9%  of  the  same  nucleotides  on  the  genome.  Thus,  if  we  know  only  a  few
complete  genomes  from  the  species,  then  we  already  possess  an  important  key
 to  unlocking  the  species  genome.

Determining  an  organism's  complete  genome  (called

genome  sequencing)  forms  a  central  task  of
bioinformatics.  Unfortunately,  we  still  don't  possess  the
microscope  technology  to  zoom  into  the  nucleotide  level
and  determine  the  sequence  of  a  genome's  nucleotides,
one  at  a  time.  However,  researchers  can  apply  chemical
methods  to  generate  and  identify  much  smaller  snippets  of
DNA,  called  reads.

After  obtaining  a  large  collection  of  reads  from  multiple Figure  1.  Fragment  Assembly  works
copies  of  the  same  genome,  the  aim  is  to  reconstruct  the by  blasting  many  copies  of  the  same
genome  into  smaller,  identifiable
desired  genome  from  these  small  pieces  of  DNA;;  this reads,  which  are  then  used  to
process  is  called  fragment  assembly  (see  Figure  1). computationally  assemble  one  copy
of  the  genome.

Problem

For  a  collection  of  strings,  a  larger  string  containing  every  one  of  the  smaller  strings  as  a  substring  is
called  a  superstring.

By  the  assumption  of  parsimony,  a  shortest  possible  superstring  over  a  collection  of  reads  serves  as  a
candidate  chromosome.

Given:  At  most  50  DNA  strings  whose  length  does  not  exceed  1  kbp  in  FASTA  format  (which
represent  reads  deriving  from  the  same  strand  of  a  single  linear  chromosome).

The  dataset  is  guaranteed  to  satisfy  the  following  condition:  there  exists  a  unique  way  to  reconstruct
the  entire  chromosome  from  these  reads  by  gluing  together  pairs  of  reads  that  overlap  by  more  than
half  their  length.

Return:  A  shortest  superstring  containing  all  the  given  strings  (thus  corresponding  to  a
reconstructed  chromosome).

Sample  Dataset

>Rosalind_56
ATTAGACCTG
>Rosalind_57
CCTGCCGGAA
>Rosalind_58
AGACCTGCCG
>Rosalind_59
GCCGGAATAC

Sample  Output

ATTAGACCTGCCGGAATAC

Extra  Information

Although  the  goal  of  fragment  assembly  is  to  produce  an  entire  genome,  in  practice  it  is  only
 possible  to  construct  several  contiguous  portions  of  each  chromosome,  called  contigs.
Furthermore,  the  assumption  made  above  that  reads  all  derive  from  the  same  strand  is  also
practically  unrealistic;;  in  reality,  researchers  will  not  know  the  strand  of  DNA  from  which  a  given
read  has  been  sequenced.

Problem  32
Perfect  Matchings  and  RNA  Secondary  Structures

Introduction  to  RNA  Folding

Because  RNA  is  single-­stranded,  you

may  have  wondered  if  the  cytosine  and
guanine  bases  bond  to  each  other  like  in
DNA.  The  answer  is  yes,  as  do  adenine
and  uracil,  and  the  resulting  base  pairs  define  the
secondary  structure  of  the  RNA  molecule;;  recall  that  its
primary  structure  is  just  the  order  of  its  bases.

In  the  greater  three-­dimensional  world,  the  base  pairing

interactions  of  an  RNA  molecule  cause  it  to  twist  around Figure  1.  A  hairpin  loop  is  formed
when  consecutive  elements  from
on  itself  in  a  process  called  RNA  folding.  When  two two  different  regions  of  an  RNA
complementary  intervals  of  bases  located  close  to  each molecule  base  pair.
other  on  the  strand  bond  to  each  other,  they  form  a
structure  called  a  hairpin  loop  (or  stem  loop),  shown  in  Figure  1.

The  same  RNA  molecule  may  base  pair  differently  at  different  points  in  time,  thus  adopting  many
different  secondary  structures.  Our  eventual  goal  is  to  classify  which  of  these  structures  are
practically  feasible,  and  which  are  not.  To  this  end,  we  will  ask  natural  combinatorial  questions
about  the  number  of  possible  different  RNA  secondary  structures.  In  this  problem,  we  will  first
consider  the  (impractical)  simplified  case  in  which  every  nucleotide  forms  part  of  a  base  pair  in  the
RNA  molecule.

Problem

A  matching  in  a  graph    is  a  collection  of  edges  of    for

which  no  node  belongs  to  more  than  one  edge  in  the
collection.  See  Figure  2  for  examples  of  matchings.  If  
contains  an  even  number  of  nodes  (say   ),  then  a  matching Figure  2.  Three  matchings
(highlighted  in  red)  shown  in  three
on    is  perfect  if  it  contains    edges,  which  is  clearly  the
different  graphs.
maximum  possible.  An  example  of  a  graph  containing  a
perfect  matching  is  shown  in  Figure  3.

First,  let    denote  the  complete  graph  on    labeled  nodes,  in  which  every  node  is  connected  to
every  other  node  with  an  edge,  and  let    denote  the  total  number  of  perfect  matchings  in   .  For  a
given  node   ,  there  are    ways  to  join    to  the  other  nodes  in  the  graph,  after  which  point  we
must  form  a  perfect  matching  on  the  remaining    nodes.  This  reasoning  provides  us  with  the
recurrence  relation   ;;  using  the  fact  that    is  1,  this  recurrence  relation
implies  the  closed  equation   .
 Given  an  RNA  string   ,  a  bonding  graph  for  
is  formed  as  follows.  First,  assign  each  symbol  of    to  a  node,
and  arrange  these  nodes  in  order  around  a  circle,  connecting
them  with  edges  called  adjacency  edges.  Second,  form  all
possible  edges  {A,  U}  and  {C,  G},  called  basepair  edges;;  we
will  represent  basepair  edges  with  dashed  edges,  as
illustrated  by  the  bonding  graph  in  Figure  4.

Note  that  a  matching  contained  in  the  basepair  edges  will

represent  one  possibility  for  base  pairing  interactions  in   ,  as
shown  in  Figure  5.  For  such  a  matching  to  exist,    must  have
the  same  number  of  'A's  as  'U's  and  the  same  number  of  'C's
as  'G's. Figure  3.  This  graph  contains  10
nodes;;  the  five  edges  forming  a
Given:  An  RNA  string    of  length  at  most  80  bp  having  the perfect  matching  on  these  nodes
same  number  of  occurrences  of  'A'  as  'U'  and  the  same are  highlighted  in  red.
number  of  occurrences  of  'C'  as  'G'.

Return:  The  total  possible  number  of  perfect  matchings  of

basepair  edges  in  the  bonding  graph  of   .

Sample  Dataset

>Rosalind_23
AGCUAGUCAU

Sample  Output
Figure  4.  The  bonding  graph  for  the
RNA  string  s  =  UAGCGUGAUCAC.
12

Figure  5.  A  perfect  matching  on  the

basepair  edges  is  highlighted  in  red
and  represents  a  candidate
secondary  structure  for  the  RNA
strand.

Problem  33
Partial  Permutations
 Partial  Gene  Orderings

Similar  species  will  share  many  of  the  same  genes,  possibly  with  modifications.
Thus,  we  can  compare  two  genomes  by  analyzing  the  orderings  of  their  genes,
then  inferring  which  rearrangements  have  separated  the  genes.

In  “Enumerating  Gene  Orders”,  we  used  permutations  to  model  gene  orderings.  Yet  two  genomes
will  have  evolved  along  their  own  separate  paths,  and  so  they  won't  share  all  of  the  same  genes.  As
a  result,  we  should  modify  the  notion  of  permutation  in  order  to  quantify  the  notion  of  partial  gene
orderings.

Problem

A  partial  permutation  is  an  ordering  of  only    objects  taken  from  a  collection  containing    objects
(i.e.,   ).  For  example,  one  partial  permutation  of  three  of  the  first  eight  positive  integers  is  given
by   .

The  statistic    counts  the  total  number  of  partial  permutations  of    objects  that  can  be  formed
from  a  collection  of    objects.  Note  that    is  just  the  number  of  permutations  of    objects,
which  we  found  to  be  equal  to    in  “Enumerating  Gene  Orders”.

Given:  Positive  integers    and    such  that    and   .

Return:  The  total  number  of  partial  permutations   ,  modulo  1,000,000.

Sample  Dataset

21  7

Sample  Output

51200

Problem  34
Introduction  to  Random  Strings

Modeling  Random  Genomes

We  already  know  that  the  genome  is  not  just  a  random  strand  of  nucleotides;;
recall  from  “Finding  a  Motif  in  DNA”  that  motifs  recur  commonly  across  individuals
and  species.  If  a  DNA  motif  occurs  in  many  different  organisms,  then  chances  are
good  that  it  serves  an  important  function.

At  the  same  time,  if  you  form  a  long  enough  DNA  string,  then  you  should  theoretically  be  able  to
 locate  every  possible  short  substring  in  the  string.  And  genomes  are  very  long;;  the  human  genome
contains  about  3.2  billion  base  pairs.  As  a  result,  when  analyzing  an  unknown  piece  of  DNA,  we
should  try  to  ensure  that  a  motif  does  not  occur  out  of  random  chance.

To  conclude  whether  motifs  are  random  or  not,  we  need  to  quantify  the  likelihood  of  finding  a  given
motif  randomly.  If  a  motif  occurs  randomly  with  high  probability,  then  how  can  we  really  compare
two  organisms  to  begin  with?  In  other  words,  all  very  short  DNA  strings  will  appear  randomly  in  a
genome,  and  very  few  long  strings  will  appear;;  what  is  the  critical  motif  length  at  which  we  can
throw  out  random  chance  and  conclude  that  a  motif  appears  in  a  genome  for  a  reason?

In  this  problem,  our  first  step  toward  understanding  random  occurrences  of  strings  is  to  form  a
simple  model  for  constructing  genomes  randomly.  We  will  then  apply  this  model  to  a  somewhat
simplified  exercise:  calculating  the  probability  of  a  given  motif  occurring  randomly  at  a  fixed  location
in  the  genome.

Problem

An  array  is  a  structure  containing  an  ordered  collection  of

objects  (numbers,  strings,  other  arrays,  etc.).  We  let  
denote  the   -­th  value  in  array   .  You  may  like  to  think  of  an
array  as  simply  a  matrix  having  only  one  row.

A  random  string  is  constructed  so  that  the  probability  of Figure  1.  The  graph  of  the  common
logarithm  function  of  x.  For  a  given  x-­
choosing  each  subsequent  symbol  is  based  on  a  fixed value,  the  corresponding  y-­value  is
underlying  symbol  frequency. the  exponent  to  which  we  must  raise
10  to  obtain  x.  Note  that  x-­values
GC-­content  offers  us  natural  symbol  frequencies  for between  0  and  1  get  mapped  to  y-­
constructing  random  DNA  strings.  If  the  GC-­content  is   ,  then values  between  -­infinity  and  0.
we  set  the  symbol  frequencies  of  C  and  G  equal  to    and  the
symbol  frequencies  of  A  and  T  equal  to   .  For  example,  if  the  GC-­content  is  40%,  then  as  we
construct  the  string,  the  next  symbol  is  'G'/'C'  with  probability  0.2,  and  the  next  symbol  is  'A'/'T'  with
probability  0.3.

In  practice,  many  probabilities  wind  up  being  very  small.  In  order  to  work  with  small  probabilities,  we
may  plug  them  into  a  function  that  "blows  them  up"  for  the  sake  of  comparison.  Specifically,  the
common  logarithm  of    (defined  for    and  denoted   )  is  the  exponent  to  which  we
must  raise  10  to  obtain   .

See  Figure  1  for  a  graph  of  the  common  logarithm  function   .  In  this  graph,  we  can  see
that  the  logarithm  of   -­values  between  0  and  1  always  winds  up  mapping  to   -­values  between  
and  0:   -­values  near  0  have  logarithms  close  to   ,  and   -­values  close  to  1  have  logarithms  close
to   .  Thus,  we  will  select  the  common  logarithm  as  our  function  to  "blow  up"  small  probability  values
for  comparison.

Given:  A  DNA  string    of  length  at  most  100  bp  and  an  array    containing  at  most  20  numbers
between  0  and  1.

Return:  An  array    having  the  same  length  as    in  which    represents  the  common  logarithm
of  the  probability  that  a  random  string  constructed  with  the  GC-­content  found  in    will  match  
exactly.

Sample  Dataset

ACGATACAA
0.129  0.287  0.423  0.476  0.641  0.742  0.783
 Sample  Output

-­‐5.737  -­‐5.217  -­‐5.263  -­‐5.360  -­‐5.958  -­‐6.628  -­‐7.009

Hint

One  property  of  the  logarithm  function  is  that  for  any  positive  numbers    and   ,  
.

Problem  35
Enumerating  Oriented  Gene  Orderings

Synteny  Blocks  Have  Orientations

In  “Enumerating  Gene  Orders”,  we  introduced  synteny  blocks  for  two  different
species,  which  are  very  similar  areas  of  two  species  genomes  that  have  been
flipped  and  moved  around  by  rearrangements.  In  that  problem,  we  used  the
permutation  to  model  the  order  of  synteny  blocks  on  a  single  chromosome.

However,  each  strand  of  a  DNA  molecule  has  an  orientation  (as  RNA  transcription  only  occurs  in
one  direction),  and  so  to  more  prudently  model  chromosomes  using  synteny  blocks,  we  should
provide  each  block  with  an  orientation  to  indicate  the  strand  on  which  it  is  located.  Adding
orientations  to  synteny  blocks  requires  us  to  expand  our  notion  of  permutation  so  that  each  index
in  the  permutation  has  its  own  orientation.

Problem

A  signed  permutation  of  length    is  some  ordering  of  the  positive  integers    in  which
each  integer  is  then  provided  with  either  a  positive  or  negative  sign  (for  the  sake  of  simplicity,  we  omit
the  positive  sign).  For  example,    is  a  signed  permutation  of  length   .

Given:  A  positive  integer   .

Return:  The  total  number  of  signed  permutations  of  length   ,  followed  by  a  list  of  all  such
permutations  (you  may  list  the  signed  permutations  in  any  order).

Sample  Dataset

Sample  Output

8
 -­‐1  -­‐2
-­‐1  2
1  -­‐2
1  2
-­‐2  -­‐1
-­‐2  1
2  -­‐1
2  1

Problem  36
Finding  a  Spliced  Motif

Motifs  Are  Rarely  Contiguous

In  “Finding  a  Motif  in  DNA”,  we  searched  for  occurrences  of  a  motif  as  a  substring
of  a  larger  database  genetic  string.  However,  because  a  DNA  strand  coding  for  a
protein  is  often  interspersed  with  introns  (see  “RNA  Splicing”),  we  need  a  way  to
recognize  a  motif  that  has  been  chopped  up  into  pieces  along  a  chromosome.

Problem

A  subsequence  of  a  string  is  a  collection  of  symbols  contained  in  order  (though  not  necessarily
contiguously)  in  the  string  (e.g.,  ACG  is  a  subsequence  of  TATGCTAAGATC).  The  indices  of  a
subsequence  are  the  positions  in  the  string  at  which  the  symbols  of  the  subsequence  appear;;  thus,
the  indices  of  ACG  in  TATGCTAAGATC  can  be  represented  by  (2,  5,  9).

As  a  substring  can  have  multiple  locations,  a  subsequence  can  have  multiple  collections  of  indices,
and  the  same  index  can  be  reused  in  more  than  one  appearance  of  the  subsequence;;  for  example,
ACG  is  a  subsequence  of  AACCGGTT  in  8  different  ways.

Given:  Two  DNA  strings    and    (each  of  length  at  most  1  kbp)  in  FASTA  format.
Return:  One  collection  of  indices  of    in  which  the  symbols  of    appear  as  a  subsequence  of   .  If
multiple  solutions  exist,  you  may  return  any  one.

Sample  Dataset

>Rosalind_14
ACGTACGTGACG
>Rosalind_18
GTA

Sample  Output

3  8  10
 Extra  Information

For  the  mathematically  inclined,  we  may  equivalently  say  that    is  a  subsequence
of    if  the  characters  of    appear  in  the  same  order  within   .  Even  more  formally,  a
subsequence  of    is  a  string   ,  where   .

Problem  37
Transitions  and  Transversions

Certain  Point  Mutations  are  More  Common

Point  mutations  occurring  in  DNA  can  be

divided  into  two  types:  transitions  and
transversions.  A  transition  substitutes
one  purine  for  another  ( )  or  one
pyrimidine  for  another  ( );;  that  is,  a  transition  does
not  change  the  structure  of  the  nucleobase.  Conversely,  a
transversion  is  the  interchange  of  a  purine  for  a  pyrimidine
base,  or  vice-­versa.  See  Figure  1.  Transitions  and
transversions  can  be  defined  analogously  for  RNA
mutations.

Because  transversions  require  a  more  drastic  change  to

the  base's  chemical  structure,  they  are  less  common  than Figure  1.  Illustration  of  transitions
transitions.  Across  the  entire  genome,  the  ratio  of and  transversions.
transitions  to  transversions  is  on  average  about  2.
However,  in  coding  regions,  this  ratio  is  typically  higher  (often  exceeding  3)  because  a  transition
appearing  in  coding  regions  happens  to  be  less  likely  to  change  the  encoded  amino  acid,
particularly  when  the  substituted  base  is  the  third  member  of  a  codon  (feel  free  to  verify  this  fact
using  the  DNA  codon  table).  Such  a  substitution,  in  which  the  organism's  protein  makeup  is
unaffected,  is  known  as  a  silent  substitution.

Because  of  its  potential  for  identifying  coding  DNA,  the  ratio  of  transitions  to  transversions  between
two  strands  of  DNA  offers  a  quick  and  useful  statistic  for  analyzing  genomes.

Problem

For  DNA  strings    and    having  the  same  length,  their  transition/transversion  ratio    is
the  ratio  of  the  total  number  of  transitions  to  the  total  number  of  transversions,  where  symbol
substitutions  are  inferred  from  mismatched  corresponding  symbols  as  when  calculating  Hamming
distance  (see  “Counting  Point  Mutations”).

Given:  Two  DNA  strings    and    of  equal  length  (at  most  1  kbp).

Return:  The  transition/transversion  ratio   .

 Sample  Dataset

>Rosalind_0209
GCAACGCACAACGAAAACCCTTAGGGACTGGATTATTTCGTGATCGTTGTAGTTATTGGA
AGTACGGGCATCAACCCAGTT
>Rosalind_2200
TTATCTGACAAAGAAAGCCGTCAACGGCTGGATAATTTCGCGATCGTGCTGGTTACTGGC
GGTACGAGTGTTCCTTTGGGT

Sample  Output

1.21428571429

Problem  38
Completing  a  Tree

The  Tree  of  Life

"As  buds  give  rise  by  growth  to  fresh

buds,  and  these,  if  vigorous,  branch  out
and  overtop  on  all  sides  many  a  feebler
branch,  so  by  generation  I  believe  it  has
been  with  the  great  Tree  of  Life,  which  fills  with  its  dead
and  broken  branches  the  crust  of  the  earth,  and  covers  the
surface  with  its  ever-­branching  and  beautiful  ramifications."
Charles  Darwin,  The  Origin  of  Species

A  century  and  a  half  has  passed  since  the  publication  of Figure  1.  A  phylogeny  illustrating
proposed  evolutionary  relationships
Darwin's  magnum  opus,  and  yet  the  construction  of  a among  the  three  domains  of  life:
single  Tree  of  Life  uniting  life  on  Earth  still  has  not  been Bacteria,  Archaea,  and  Eukaryota.
completed,  with  perhaps  as  many  as  90%  of  all  living
species  not  yet  catalogued  (although  a  beautiful  interactive  animation  has  been  produced  by
OneZoom).

To  get  an  insight  about  state-­of-­art  attempts  to  build  this  tree,  you  may  take  a  look  at  the  Tree  of
Life  Web  Project  -­  collaborative  effort  of  biologists  from  around  the  world  to  combine  information
about  diversity  of  life  on  Earth.  It  is  a  peer-­reviewed  ongoing  project  started  in  1995,  now  it  holds
more  than  10,000  pages  with  characteristics  of  different  groups  of  organisms  and  their  evolutionary
history,  and  tree  still  grows.

Instead  of  trying  to  construct  the  entire  Tree  of  Life  all  at  once,  we  often  wish  to  form  a  simpler  tree
in  which  a  collection  of  species  have  been  clumped  together  for  the  sake  of  simplicity;;  such  a
group  is  called  a  taxon  (pl.  taxa).  For  a  given  collection  of  taxa,  a  phylogeny  is  a  treelike  diagram
that  best  represents  the  evolutionary  connections  between  the  taxa:  the  construction  of  a  particular
phylogeny  depends  on  our  specific  assumptions  regarding  how  these  evolutionary  relationships
 should  be  interpreted.  See  Figure  1.

Problem

An  undirected  graph  is  connected  if  there  is  a  path

connecting  any  two  nodes.  A  tree  is  a  connected  (undirected)
graph  containing  no  cycles;;  this  definition  forces  the  tree  to
have  a  branching  structure  organized  around  a  central  core  of
nodes,  just  like  its  living  counterpart.  See  Figure  2.

We  have  already  grown  familiar  with  trees  in  “Mendel's  First

Law”,  where  we  introduced  the  probability  tree  diagram  to
visualize  the  outcomes  of  a  random  variable.

In  the  creation  of  a  phylogeny,  taxa  are  encoded  by  the  tree's
leaves,  or  nodes  having  degree  1.  A  node  of  a  tree  having
degree  larger  than  1  is  called  an  internal  node. Figure  2.  A  labeled  tree  with  6
vertices  and  5  edges.
Given:  A  positive  integer    ( )  and  an  adjacency
list  corresponding  to  a  graph  on    nodes  that  contains  no
cycles.

Return:  The  minimum  number  of  edges  that  can  be  added  to  the  graph  to  produce  a  tree.

Sample  Dataset

10
1  2
2  8
4  10
5  9
6  10
7  9

Sample  Output

Extra  Information

After  solving  this  problem,  a  standard  mathematical  exercise  for  the  technically  minded  is  to  verify
that  every  tree  having  2  or  more  nodes  must  contain  at  least  two  leaves.

Problem  39
Catalan  Numbers  and  RNA  Secondary  Structures
 The  Human  Knot

You  may  have  had  the  misfortune  to

participate  in  a  team-­building  event  that
featured  the  "human  knot,"  in  which
everyone  joins  hands  with  two  other
people,  and  the  group  must  undo  the  giant  knot  of  arms
without  letting  go  (see  Figure  1).

Let's  consider  a  simplified  version  of  the  human  knot.  Say

that  we  have  an  even  number  of  people  at  a  party  who  are
standing  in  a  circle,  and  they  pair  off  and  shake  hands  at Figure  1.  Knot  fun.  Courtesy  El  Photo
the  same  time.  One  combinatorial  question  at  hand  asks Studio.
us  to  count  the  total  number  of  ways  that  the  guests  can
shake  hands  without  any  two  pairs  interfering  with  each
other  by  crossing  arms.

This  silly  little  counting  problem  is  actually  an  excellent

analogy  for  RNA  folding.  In  practice,  base  pairing  can
occur  anywhere  along  the  RNA  molecule,  but  the
secondary  structure  of  RNA  often  forbids  base  pairs
crossing  over  each  other,  which  forms  a  structure  called  a Figure  2.  This  pseudoknot  was
formed  when  bonding  occurred  at
pseudoknot  (see  Figure  2)).  Pseudoknots  are  not the  endpoints  of  overlapping
technically  knots,  but  they  nevertheless  cause  RNA  to  fold intervals  [1,3]  and  [2,  4].
over  itself.

Forbidding  pseudoknots  offers  an  interesting  wrinkle  to  the  problem  of  counting  potential  RNA
secondary  structures  that  we  started  working  with  in  “Perfect  Matchings  and  RNA  Secondary
Structures”,  in  which  every  possible  nucleotide  of  a  strand  of  RNA  must  base  pair  with  another
nucleotide.

Problem

A  matching  in  a  graph  is  noncrossing  if  none  of  its  edges
cross  each  other.  If  we  assume  that  the    nodes  of  this  graph
are  arranged  around  a  circle,  and  if  we  label  these  nodes  with
positive  integers  between  1  and   ,  then  a  matching  is
noncrossing  as  long  as  there  are  not  edges    and  
such  that   .
Figure  3.  The  only  two  noncrossing
A  noncrossing  matching  of  basepair  edges  in  the  bonding perfect  matchings  of  basepair
graph  corresponding  to  an  RNA  string  will  correspond  to  a edges  (shown  in  red)  for  the  RNA
string  UAGCGUGAUCAC.
possible  secondary  structure  of  the  underlying  RNA  strand
that  lacks  pseudoknots,  as  shown  in  Figure  3.

In  this  problem,  we  will  consider  counting  noncrossing  perfect

matchings  of  basepair  edges.  As  a  motivating  example  of  how
to  count  noncrossing  perfect  matchings,  let    denote  the
number  of  noncrossing  perfect  matchings  in  the  complete
graph   .  After  setting   ,  we  can  see  that    should
equal  1  as  well.  As  for  the  case  of  a  general   ,  say  that  the
nodes  of    are  labeled  with  the  positive  integers  from  1  to  
.  We  can  join  node  1  to  any  of  the  remaining  
Figure  4.  This  figure  shows  all
nodes;;  yet  once  we  have  chosen  this  node  (say   ),  we possible  noncrossing  perfect
cannot  add  another  edge  to  the  matching  that  crosses  the matchings  in  complete  graphs  on  2,
 edge   .  As  a  result,  we  must  match  all  the  edges  on 4,  6,  and  8  nodes;;  the  total  number
of  such  matchings  are  1,  2,  5,  and
one  side  of    to  each  other.  This  requirement  forces   14,  respectively.  Courtesy  Tom
to  be  even,  so  that  we  can  write    for  some  positive Davis.
integer   .

There  are    nodes  on  one  side  of    and    nodes  on  the  other  side  of   ,  so
that  in  turn  there  will  be    different  ways  of  forming  a  perfect  matching  on  the  remaining
nodes  of   .  If  we  let    vary  over  all  possible    choices  of  even  numbers  between  1  and   ,
then  we  obtain  the  recurrence  relation   .  The  resulting  numbers    counting
noncrossing  perfect  matchings  in    are  called  the  Catalan  numbers,  and  they  appear  in  a  huge
number  of  other  settings.  See  Figure  4  for  an  illustration  counting  the  first  four  Catalan  numbers.

Given:  An  RNA  string    having  the  same  number  of  occurrences  of  'A'  as  'U'  and  the  same  number
of  occurrences  of  'C'  as  'G'.  The  length  of  the  string  is  at  most  300  bp.

Return:  The  total  number  of  noncrossing  perfect  matchings  of  basepair  edges  in  the  bonding  graph
of   ,  modulo  1,000,000.

Sample  Dataset

>Rosalind_57
AUAU

Sample  Output

Hint

Write  a  function  that  counts  Catalan  numbers  via  dynamic  programming.  How  can  we  modify  this
function  to  apply  to  our  given  problem?

Problem  40
Error  Correction  in  Reads

Genome  Sequencing  Isn't  Perfect

In  “Genome  Assembly  as  Shortest  Superstring”,  we  introduce  the  problem  of
assembling  a  genome  from  a  collection  of  reads.  Even  though  genome
sequencing  is  a  multi-­billion  dollar  enterprise,  sequencing  machines  that  identify
reads  still  produce  errors  a  substantial  percentage  of  the  time.  To  make  matters
worse,  these  errors  are  unpredictable;;  it  is  difficult  to  determine  if  the  machine  has  made  an  error,
let  alone  where  in  the  read  the  error  has  occurred.  For  this  reason,  error  correction  in  reads  is
typically  a  vital  first  step  in  genome  assembly.
 Problem

As  is  the  case  with  point  mutations,  the  most  common  type  of  sequencing  error  occurs  when  a  single
nucleotide  from  a  read  is  interpreted  incorrectly.

Given:  A  collection  of  up  to  1000  reads  of  equal  length  (at  most  50  bp)  in  FASTA  format.  Some  of
these  reads  were  generated  with  a  single-­nucleotide  error.  For  each  read    in  the  dataset,  one  of  the
following  applies:

 was  correctly  sequenced  and  appears  in  the  dataset  at  least  twice  (possibly  as  a  reverse
complement);;
 is  incorrect,  it  appears  in  the  dataset  exactly  once,  and  its  Hamming  distance  is  1  with  respect
to  exactly  one  correct  read  in  the  dataset  (or  its  reverse  complement).

Return:  A  list  of  all  corrections  in  the  form  "[old  read]-­>[new  read]".  (Each  correction  must  be  a
single  symbol  substitution,  and  you  may  return  the  corrections  in  any  order.)

Sample  Dataset

>Rosalind_52
TCATC
>Rosalind_44
TTCAT
>Rosalind_68
TCATC
>Rosalind_28
TGAAA
>Rosalind_95
GAGGA
>Rosalind_66
TTTCA
>Rosalind_33
ATCAA
>Rosalind_21
TTGAT
>Rosalind_18
TTTCC

Sample  Output

TTCAT-­‐>TTGAT
GAGGA-­‐>GATGA
TTTCC-­‐>TTTCA

Problem  41
Counting  Phylogenetic  Ancestors
 Culling  the  Forest

In  “Completing  a  Tree”,  we  introduced  the

tree  for  the  purposes  of  constructing
phylogenies.  Yet  the  definition  of  tree  as
a  connected  graph  with  no  cycles
produces  a  huge  class  of  different  graphs,  from  simple Figure  1.  Trees  come  in  lots  of
paths  and  star-­like  graphs  to  more  familiar  arboreal different  shapes.
structures  (see  Figure  1).  Which  of  these  graphs  are
appropriate  for  phylogenetic  study?

Modern  evolutionary  theory  (beginning  with  Darwin)  indicates  that  the  only  way  a  new  species  can
be  created  is  if  it  splits  off  from  an  existing  species  after  a  population  is  isolated  for  an  extended
period  of  time.  This  model  of  species  evolution  implies  a  very  specific  type  of  phylogeny,  in  which
internal  nodes  represent  branching  points  of  evolution  where  an  ancestor  species  either  evolved  into
a  new  species  or  split  into  two  new  species:  therefore,  one  edge  of  this  internal  node  therefore
connects  the  node  to  its  most  recent  ancestor,  whereas  one  or  two  new  edges  connect  it  to  its
immediate  descendants.  This  framework  offers  a  much  clearer  notion  of  how  to  characterize
phylogenies.

Problem

A  binary  tree  is  a  tree  in  which  each  node  has  degree  equal  to  at  most  3.  The  binary  tree  will  be  our
main  tool  in  the  construction  of  phylogenies.

A  rooted  tree  is  a  tree  in  which  one  node  (the  root)  is  set  aside  to  serve  as  the  pinnacle  of  the  tree.  A
standard  graph  theory  exercise  is  to  verify  that  for  any  two  nodes  of  a  tree,  exactly  one  path  connects
the  nodes.  In  a  rooted  tree,  every  node    will  therefore  have  a  single  parent,  or  the  unique  node  
such  that  the  path  from    to  the  root  contains   .  Any  other  node    adjacent  to    is  called  a
child  of    because    must  be  the  parent  of   ;;  note  that  a  node  may  have  multiple  children.  In  other
words,  a  rooted  tree  possesses  an  ordered  hierarchy  from  the  root  down  to  its  leaves,  and  as  a  result,
we  may  often  view  a  rooted  tree  with  undirected  edges  as  a  directed  graph  in  which  each  edge  is
oriented  from  parent  to  child.  We  should  already  be  familiar  with  this  idea;;  it's  how  the  Rosalind
problem  tree  works!

Even  though  a  binary  tree  can  include  nodes  having  degree  2,  an  unrooted  binary  tree  is  defined
more  specifically:  all  internal  nodes  have  degree  3.  In  turn,  a  rooted  binary  tree  is  such  that  only  the
root  has  degree  2  (all  other  internal  nodes  have  degree  3).

Given:  A  positive  integer    ( ).

Return:  The  number  of  internal  nodes  of  any  unrooted  binary  tree  having    leaves.

Sample  Dataset

Sample  Output

2
 Hint

In  solving  “Completing  a  Tree”,  you  may  have  formed  the  conjecture  that  a  graph  with  no  cycles  and
 nodes  is  a  tree  precisely  when  it  has    edges.  This  is  indeed  a  theorem  of  graph  theory.

Problem  42
k-­Mer  Composition

Generalizing  GC-­Content

A  length    substring  of  a  genetic  string  is

commonly  called  a  k-­mer.  A  genetic
Figure  1.  The  2-­mer  composition  of
string  of  length    can  be  seen  as TTGATTACCTTATTTGATCATTACACATTGTACGCTT
composed  of    overlapping  k-­
mers.  The  k-­mer  composition  of  a  genetic  string  encodes  the  number  of  times  that  each  possible
k-­mer  occurs  in  the  string.  See  Figure  1.  The  1-­mer  composition  is  a  generalization  of  the  GC-­
content  of  a  strand  of  DNA,  and  the  2-­mer,  3-­mer,  and  4-­mer  compositions  of  a  DNA  string  are  also
commonly  known  as  its  di-­nucleotide,  tri-­nucleotide,  and  tetra-­nucleotide  compositions.

The  biological  significance  of  k-­mer  composition  is  manyfold.  GC-­content  is  helpful  not  only  in
helping  to  identify  a  piece  of  unknown  DNA  (see  “Computing  GC  Content”),  but  also  because  a
genomic  region  having  high  GC-­content  compared  to  the  rest  of  the  genome  signals  that  it  may
belong  to  an  exon.  Analyzing  k-­mer  composition  is  vital  to  fragment  assembly  as  well.

In  “Computing  GC  Content”,  we  also  drew  an  analogy  between  analyzing  the  frequency  of
characters  and  identifying  the  underlying  language.  For  larger  values  of   ,  the  k-­mer  composition
offers  a  more  robust  fingerprint  of  a  string's  identity  because  it  offers  an  analysis  on  the  scale  of
substrings  (i.e.,  words)  instead  of  that  of  single  symbols.  As  a  basis  of  comparison,  in  language
analysis,  the  k-­mer  composition  of  a  text  can  be  used  not  only  to  pin  down  the  language,  but  also
often  the  author.

Problem

For  a  fixed  positive  integer   ,  order  all  possible  k-­mers  taken  from  an  underlying  alphabet
lexicographically.

Then  the  k-­mer  composition  of  a  string    can  be  represented  by  an  array    for  which    denotes
the  number  of  times  that  the   th  k-­mer  (with  respect  to  the  lexicographic  order)  appears  in   .

Given:  A  DNA  string    in  FASTA  format  (having  length  at  most  100  kbp).
Return:  The  4-­mer  composition  of   .

Sample  Dataset

>Rosalind_6431
CTTCGAAAGTTTGGGCCGAGTCTTACAGTCGGTCTTGAAGCAAAGTAACGAACTCCACGG
 CCCTGACTACCGAACCAGTTGTGAGTACTCAACTGGGTGAGAGTGCAGTCCCTATTGAGT
TTCCGAGACTCACCGGGATTTTCGATCCAGCCTCAGTCCAGTCTTGTGGCCAACTCACCA
AATGACGTTGGAATATCCCTGTCTAGCTCACGCAGTACTTAGTAAGAGGTCGCTGCAGCG
GGGCAAGGAGATCGGAAAATGTGCTCTATATGCGACTAAAGCTCCTAACTTACACGTAGA
CTTGCCCGTGTTAAAAACTCGGCTCACATGCTGTCTGCGGCTGGCTGTATACAGTATCTA
CCTAATACCCTTCAGTTCGCCGCACAAAAGCTGGGAGTTACCGCGGAAATCACAG

Sample  Output

4  1  4  3  0  1  1  5  1  3  1  2  2  1  2  0  1  1  3  1  2  1  3  1  1  1  1  2  2  5  1  3  0  2  2  1
 1  1  1  3  1  0  0  1  5  5  1  5  0  2  0  2  1  2  1  1  1  2  0  1  0  0  1  1  3  2  1  0  3  2  3  
0  0  2  0  8  0  0  1  0  2  1  3  0  0  0  1  4  3  2  1  1  3  1  2  1  3  1  2  1  2  1  1  1  2  3  2
 1  1  0  1  1  3  2  1  2  6  2  1  1  1  2  3  3  3  2  3  0  3  2  1  1  0  0  1  4  3  0  1  5  0  2  
0  1  2  1  3  0  1  2  2  1  1  0  3  0  0  4  5  0  3  0  2  1  1  3  0  3  2  2  1  1  0  2  1  0  2  2
 1  2  0  2  2  5  2  2  1  1  2  1  2  2  2  2  1  1  3  4  0  2  1  1  0  1  2  2  1  1  1  5  2  0  3  
2  1  1  2  2  3  0  3  0  1  3  1  2  3  0  2  1  2  2  1  2  3  0  1  2  3  1  1  3  1  0  1  1  3  0  2
 1  2  2  0  2  1  1

Problem  43
Speeding  Up  Motif  Finding

Shortening  the  Motif  Search

In  “Finding  a  Motif  in  DNA”,  we  discussed  the  problem  of  searching  a  genome  for
a  known  motif.  Because  of  the  large  scale  of  eukaryotic  genomes,  we  need  to
accomplish  this  computational  task  as  efficiently  as  possible.

The  standard  method  for  locating  one  string    as  a  substring  of  another  string    (and  perhaps  one
you  implemented  in  “Finding  a  Motif  in  DNA”)  is  to  move  a  sliding  window  across  the  larger  string,
at  each  step  starting  at    and  matching  subsequent  symbols  of    to  symbols  of   .  After  we  have
located  a  match  or  mismatch,  we  then  shift  the  window  backwards  to  begin  searching  at   .

The  potential  weakness  of  this  method  is  as  follows:  say  we  have  matched  100  symbols  of    to  
before  reaching  a  mismatch.  The  window-­sliding  method  would  then  move  back  99  symbols  of  
and  start  comparing    to   ;;  can  we  avoid  some  of  this  sliding?

For  example,  say  that  we  are  looking  for    in  

.  From    to   ,  we  have  matched  seven
symbols  of   ,  and  yet    =  G  produces  a  mismatch  with    =  T.  We  don't  need  to  go  all  the
way  back  to    and  start  matching  with    because   ,   ,  and    are  all
different  from   .  What  about   ?  Because   ,  the  previous
mismatch  of    and    guarantees  the  same  mismatch  with    and   .
Following  this  analysis,  we  may  advance  directly  to    and  continue  sliding  our  window,  without
ever  having  to  move  it  backward.

This  method  can  be  generalized  to  form  the  framework  behind  the  Knuth-­Morris-­Pratt  algorithm
(KMP),  which  was  published  in  1977  and  offers  an  efficiency  boost  for  determining  whether  a  given
 motif  can  be  located  within  a  larger  string.

Problem

A  prefix  of  a  length    string    is  a  substring   ;;  a  suffix  of    is  a  substring   .

The  failure  array  of    is  an  array    of  length    for  which    is  the  length  of  the  longest  substring  
 that  is  equal  to  some  prefix   ,  where    cannot  equal    (otherwise,  
would  always  equal   ).  By  convention,   .

Given:  A  DNA  string    (of  length  at  most  100  kbp)  in  FASTA  format.
Return:  The  failure  array  of   .

Sample  Dataset

>Rosalind_87
CAGCATGGTATCACAGCAGAG

Sample  Output

0  0  0  1  2  0  0  0  0  0  0  1  2  1  2  3  4  5  3  0  0

Extra  Information

If  you  would  like  a  more  precise  technical  explanation  of  the  Knuth-­Morris-­Pratt  algorithm,  please
take  a  look  at  this  site

Problem  44
Finding  a  Shared  Spliced  Motif

Locating  Motifs  Despite  Introns

In  “Finding  a  Shared  Motif”,  we  discussed  searching  through  a  database

containing  multiple  genetic  strings  to  find  a  longest  common  substring  of  these
strings,  which  served  as  a  motif  shared  by  the  two  strings.  However,  as  we  saw  in
“RNA  Splicing”,  coding  regions  of  DNA  are  often  interspersed  by  introns  that  do
not  code  for  proteins.

We  therefore  need  to  locate  shared  motifs  that  are  separated  across  exons,  which  means  that  the
motifs  are  not  required  to  be  contiguous.  To  model  this  situation,  we  need  to  enlist  subsequences.

Problem
 A  string    is  a  common  subsequence  of  strings    and    if  the  symbols  of    appear  in  order  as  a
subsequence  of  both    and   .  For  example,  "ACTG"  is  a  common  subsequence  of  "AACCTTGG"  and
"ACACTGTGA".

Analogously  to  the  definition  of  longest  common  substring,    is  a  longest  common  subsequence  of  
 and    if  there  does  not  exist  a  longer  common  subsequence  of  the  two  strings.  Continuing  our  above
example,  "ACCTTG"  is  a  longest  common  subsequence  of  "AACCTTGG"  and  "ACACTGTGA",  as  is
"AACTGG".

Given:  Two  DNA  strings    and    (each  having  length  at  most  1  kbp)  in  FASTA  format.
Return:  A  longest  common  subsequence  of    and   .  (If  more  than  one  solution  exists,  you  may
return  any  one.)

Sample  Dataset

>Rosalind_23
AACCTTGG
>Rosalind_64
ACACTGTGA

Sample  Output

AACTGG

Problem  45
Ordering  Strings  of  Varying  Length  Lexicographically

Organizing  Strings  of  Different  Lengths

In  “Enumerating  k-­mers  Lexicographically”,  we  introduced  the  lexicographic  order

for  strings  of  the  same  length  constructed  from  some  ordered  underlying  alphabet.
However,  our  experience  with  dictionaries  suggests  that  we  should  be  able  to
order  strings  of  different  lengths  just  as  easily.  That  is,  we  already  have  an
intuitive  sense  that  "APPLE"  comes  before  "APPLET",  which  comes  before  "ARTS,"  and  so  we
should  be  able  to  apply  this  intuition  toward  cataloguing  genetic  strings  of  varying  lengths.

Problem

Say  that  we  have  strings    and    with   .  Consider  the  substring
.  We  have  two  cases:

1.   If   ,  then  we  set    because    is  shorter  than    (e.g.,   ).

2.   Otherwise,   .  We  define    if    and  define    if    (e.g.,  
 because   ).
 Given:  A  permutation  of  at  most  12  symbols  defining  an  ordered  alphabet    and  a  positive  integer  
 ( ).

Return:  All  strings  of  length  at  most    formed  from   ,  ordered  lexicographically.  (Note:  As  in
“Enumerating  k-­mers  Lexicographically”,  alphabet  order  is  based  on  the  order  in  which  the  symbols  are
given.)

Sample  Dataset

D  N  A
3

Sample  Output

D
DD
DDD
DDN
DDA
DN
DND
DNN
DNA
DA
DAD
DAN
DAA
N
ND
NDD
NDN
NDA
NN
NND
NNN
NNA
NA
NAD
NAN
NAA
A
AD
ADD
ADN
ADA
AN
AND
ANN
ANA
AA
AAD
AAN
 AAA

Extra  Information

We  can  combine  conditions  (1)  and  (2)  above  into  a  single  condition  by  adding  a  blank  character  
to  the  beginning  of  our  ordered  alphabet.  Then,  if    is  shorter  than   ,  we  simply  add  as  many
instances  of    as  necessary  to  make    and    the  same  length.

Problem  46
Maximum  Matchings  and  RNA  Secondary  Structures

Breaking  the  Bonds

In  “Perfect  Matchings  and  RNA  Secondary  Structures”,  we  considered  a  problem

that  required  us  to  assume  that  every  possible  nucleotide  is  involved  in  base
pairing  to  induce  an  RNA  secondary  structure.  Yet  the  only  way  this  could  occur
is  if  the  frequency  of  adenine  in  our  RNA  strand  is  equal  to  the  frequency  of  uracil
and  if  the  same  holds  for  guanine  and  cytosine.

We  will  therefore  begin  to  explore  ways  of  counting  secondary  structures  in  which  this  condition  is
not  required.  A  more  general  combinatorial  problem  will  ask  instead  for  the  total  number  of
secondary  structures  of  a  strand  having  a  maximum  possible  number  of  base  pairs.

Problem

The  graph  theoretical  analogue  of  the  quandary  stated  in  the
introduction  above  is  that  if  we  have  an  RNA  string    that  does
not  have  the  same  number  of  'C's  as  'G's  and  the  same
number  of  'A's  as  'U's,  then  the  bonding  graph  of    cannot
possibly  possess  a  perfect  matching  among  its  basepair
edges.  For  example,  see  Figure  1;;  in  fact,  most  bonding
graphs  will  not  contain  a  perfect  matching. Figure  1.  The  bonding  graph  of  s  =
UAGCGUGAUCAC  (left)  has  a
In  light  of  this  fact,  we  define  a  maximum  matching  in  a perfect  matching  of  basepair  edges,
graph  as  a  matching  containing  as  many  edges  as  possible. but  this  is  not  the  case  for  t  =
CAGCGUGAUCAC,  in  which  one
See  Figure  2  for  three  maximum  matchings  in  graphs. symbol  has  been  replaced.
A  maximum  matching  of  basepair  edges  will  correspond  to  a
way  of  forming  as  many  base  pairs  as  possible  in  an  RNA
string,  as  shown  in  Figure  3.

Given:  An  RNA  string    of  length  at  most  100. Figure  2.  A  maximum  matching
(highlighted  in  red)  is  shown  in  each
Return:  The  total  possible  number  of  maximum  matchings of  the  three  graphs  above.  Verify  that
no  other  matching  can  contain  more
of  basepair  edges  in  the  bonding  graph  of   .
edges.

Sample  Dataset
 >Rosalind_92
AUGCUUC

Sample  Output

Figure  3.  A  red  maximum  matching

of  basepair  edges  in  the  bonding
graph  for  t  =  CAGCGUGAUCAC.

Problem  47
Creating  a  Distance  Matrix

Introduction  to  Distance-­Based  Phylogeny

A  number  of  different  approaches  are  used  to  build  phylogenies,  each  one
featuring  its  own  computational  strengths  and  weaknesses.  One  of  these
measures  is  distance-­based  phylogeny,  which  constructs  a  tree  from
evolutionary  distances  calculated  between  pairs  of  taxa.

A  wide  assortment  of  different  measures  exist  for  quantifying  this  evolutionary  distance.  Once  we
have  selected  a  distance  function  and  used  it  to  calculate  the  distance  between  every  pair  of  taxa,
we  place  these  pairwise  distance  functions  into  a  table.

In  this  problem,  we  will  consider  an  evolutionary  function  based  on  Hamming  distance.  Recall  from
“Counting  Point  Mutations”  that  this  function  compares  two  homologous  strands  of  DNA  by
counting  the  minimum  possible  number  of  point  mutations  that  could  have  occurred  on  the
evolutionary  path  between  the  two  strands.

Problem

For  two  strings    and    of  equal  length,  the  p-­distance  between  them,  denoted   ,  is  the
proportion  of  corresponding  symbols  that  differ  between    and   .

For  a  general  distance  function    on    taxa    (taxa  are  often  represented  by  genetic
strings),  we  may  encode  the  distances  between  pairs  of  taxa  via  a  distance  matrix    in  which  
.

Given:  A  collection  of    ( )  DNA  strings    of  equal  length  (at  most  1  kbp).  Strings
are  given  in  FASTA  format.

Return:  The  matrix    corresponding  to  the  p-­distance    on  the  given  strings.  As  always,  note
that  your  answer  is  allowed  an  absolute  error  of  0.001.
 Sample  Dataset

>Rosalind_9499
TTTCCATTTA
>Rosalind_0942
GATTCATTTC
>Rosalind_6568
TTTCCATTTT
>Rosalind_1833
GTTCCATTTA

Sample  Output

0.00000  0.40000  0.10000  0.10000

0.40000  0.00000  0.40000  0.30000
0.10000  0.40000  0.00000  0.20000
0.10000  0.30000  0.20000  0.00000

Problem  48
Reversal  Distance

Rearrangements  Power  Large-­Scale  Genomic  Changes

Perhaps  the  most  common  type  of  genome  rearrangement  is  an  inversion,  which
flips  an  entire  interval  of  DNA  found  on  the  same  chromosome.  As  in  the  case  of
calculating  Hamming  distance  (see  “Counting  Point  Mutations”),  we  would  like  to
determine  the  minimum  number  of  inversions  that  have  occurred  on  the
evolutionary  path  between  two  chromosomes.  To  do  so,  we  will  use  the  model  introduced  in
“Enumerating  Gene  Orders”  in  which  a  chromosome  is  represented  by  a  permutation  of  its  synteny
blocks.

Problem

A  reversal  of  a  permutation  creates  a  new  permutation  by  inverting  some  interval  of  the  permutation;;  
,   ,  and    are  all  reversals  of   .  The  reversal
distance  between  two  permutations    and   ,  written   ,  is  the  minimum  number  of  reversals
required  to  transform    into    (this  assumes  that    and    have  the  same  length).

Given:  A  collection  of  at  most  5  pairs  of  permutations,  all  of  which  have  length  10.
Return:  The  reversal  distance  between  each  permutation  pair.

Sample  Dataset
 1  2  3  4  5  6  7  8  9  10
3  1  5  2  7  4  9  6  10  8

3  10  8  2  5  4  7  1  6  9
5  2  3  1  7  4  10  8  6  9

8  6  7  9  4  1  3  10  2  5
8  2  7  6  9  1  5  3  10  4

3  9  10  4  1  8  6  7  5  2
2  9  8  5  1  7  3  4  6  10

1  2  3  4  5  6  7  8  9  10
1  2  3  4  5  6  7  8  9  10

Sample  Output

9  4  5  7  0

Hint

Don't  be  afraid  to  try  an  ugly  solution.

Problem  49
Matching  Random  Motifs

More  Random  Strings

In  “Introduction  to  Random  Strings”,  we  discussed  searching  for  motifs  in  large
genomes,  in  which  random  occurrences  of  the  motif  are  possible.  Our  aim  is  to
quantify  just  how  frequently  random  motifs  occur.

One  class  of  motifs  of  interest  are  promoters,  or  regions  of  DNA  that  initiate  the  transcription  of  a
gene.  A  promoter  is  usually  located  shortly  before  the  start  of  its  gene,  and  it  contains  specific
intervals  of  DNA  that  provide  an  initial  binding  site  for  RNA  polymerase  to  initiate  transcription.
Finding  a  promoter  is  usually  the  second  step  in  gene  prediction  after  establishing  the  presence  of
an  ORF  (see  “Open  Reading  Frames”).

Unfortunately,  there  is  no  quick  rule  for  identifying  promoters.  In  Escherichia  coli,  the  promoter
contains  two  short  intervals  (TATAAT  and  TTGACA),  which  are  respectively  located  10  and  35  base
pairs  upstream  from  the  beginning  of  the  gene's  ORF.  Yet  even  these  two  short  intervals  are
consensus  strings  (see  “Consensus  and  Profile”):  they  represent  average-­case  strings  that  are  not
found  intact  in  most  promoters.  Bacterial  promoters  further  vary  in  that  some  contain  additional
intervals  used  to  bind  to  specific  proteins  or  to  change  the  intensity  of  transcription.
 Eukaryotic  promoters  are  even  more  difficult  to  characterize.  Most  have  a  TATA  box  (consensus
sequence:  TATAAA),  preceded  by  an  interval  called  a  B  recognition  element,  or  BRE.  These
elements  are  typically  located  within  40  bp  of  the  start  of  transcription.  For  that  matter,  eukaryotic
promoters  can  hold  a  larger  number  of  additional  "regulatory"  intervals,  which  can  be  found  as  far  as
several  thousand  base  pairs  upstream  of  the  gene.

Problem

Our  aim  in  this  problem  is  to  determine  the  probability  with  which  a  given  motif  (a  known  promoter,
say)  occurs  in  a  randomly  constructed  genome.  Unfortunately,  finding  this  probability  is  tricky;;  instead
of  forming  a  long  genome,  we  will  form  a  large  collection  of  smaller  random  strings  having  the  same
length  as  the  motif;;  these  smaller  strings  represent  the  genome's  substrings,  which  we  can  then  test
against  our  motif.

Given  a  probabilistic  event   ,  the  complement  of    is  the  collection    of  outcomes  not  belonging
to   .  Because    takes  place  precisely  when    does  not,  we  may  also  call    "not   ."

For  a  simple  example,  if    is  the  event  that  a  rolled  die  is  2  or  4,  then   .    is  the  event
that  the  die  is  1,  3,  5,  or  6,  and   .  In  general,  for  any  event  we  will  have  the  identity  that  
.

Given:  A  positive  integer   ,  a  number    between  0  and  1,  and  a  DNA  string    of
length  at  most  10  bp.

Return:  The  probability  that  if    random  DNA  strings  having  the  same  length  as    are  constructed
with  GC-­content    (see  “Introduction  to  Random  Strings”),  then  at  least  one  of  the  strings  equals   .
We  allow  for  the  same  random  string  to  be  created  more  than  once.

Sample  Dataset

90000  0.6
ATAGCCGA

Sample  Output

0.689

Problem  50
Counting  Subsets

Characters  and  SNPs

A  character  is  any  feature  (genetic,  physical,  etc.)  that  divides  a  collection  of
organisms  into  two  separate  groups.  One  commonly  used  genetic  character  is
the  possession  of  a  single-­nucleotide  polymorphism,  or  SNP.
 In  a  process  called  genotyping,  the  SNP  markers  taken  from  a  large  number  of  human  donors
have  been  used  very  successfully  to  catalogue  the  migration  and  differentiation  of  human
populations  over  the  last  200,000  years.  For  $199,  you  can  participate  in  National  Geographic's
Genographic  Project,  and  discover  your  own  genetic  heritage.

Whether  we  use  genetic  or  physical  characters,  we  may  think  of  a  collection  of    characters  as  a
collection  of  "ON"/"OFF"  switches.  An  organism  is  said  to  possess  a  character  in  the  "ON"
position  (although  often  the  assignment  of  "ON"/"OFF"  is  arbitrary).  Given  a  collection  of  taxa,  we
may  represent  a  character  by  the  collection  of  taxa  possessing  the  character.

Problem

A  set  is  the  mathematical  term  for  a  loose  collection  of  objects,  called  elements.  Examples  of  sets
include    and   ,  the  set  containing  all  real  numbers.  We
even  have  the  empty  set,  represented  by    or   ,  which  contains  no  elements  at  all.  Two  sets  are
equal  when  they  contain  the  same  elements.  In  other  words,  in  contrast  to  permutations,  the  ordering
of  the  elements  of  a  set  is  unimportant  (e.g.,    is  equivalent
to   ).  Sets  are  not  allowed  to  contain  duplicate  elements,
so  that    is  not  a  set.  We  have  already  used  sets  of  2
elements  to  represent  edges  from  a  graph.

A  set    is  a  subset  of    if  every  element  of    is  also  an  element  of   ,  and  we  write   .  For
example,   ,  and    is  a  subset
of  every  set  (including  itself!).

As  illustrated  in  the  biological  introduction,  we  can  use  subsets  to  represent  the  collection  of  taxa
possessing  a  character.  However,  the  number  of  applications  is  endless;;  for  example,  an  event  in
probability  can  now  be  defined  as  a  subset  of  the  set  containing  all  possible  outcomes.

Our  first  question  is  to  count  the  total  number  of  possible  subsets  of  a  given  set.

Given:  A  positive  integer    ( ).

Return:  The  total  number  of  subsets  of    modulo  1,000,000.

Sample  Dataset

Sample  Output

Hint

What  does  counting  subsets  have  to  do  with  characters  and  "ON"/"OFF"  switches?

Problem  51
Introduction  to  Alternative  Splicing

The  Baby  and  the  Bathwater

In  “RNA  Splicing”,  we  described  the

process  by  which  the  exons  are  spliced
out  from  a  molecule  of  pre-­mRNA  and
reassembled  to  yield  a  final  mRNA  for  the
purposes  of  protein  translation.

However,  the  chaining  of  exons  does  not  always  proceed  in
the  same  manner;;  alternative  splicing  describes  the  fact
that  all  the  exons  from  a  gene  are  not  necessarily  joined
together  in  order  to  produce  an  mRNA.  The  most  common Figure  1.  Alternative  splicing  induces
form  of  alternative  splicing  is  exon  skipping,  in  which different  protein  isoforms.
certain  exons  are  omitted  along  with  introns.

Alternative  splicing  serves  a  vital  evolutionary  purpose,  as  it  greatly  increases  the  number  of
different  proteins  that  can  be  translated  from  a  given  gene;;  different  proteins  produced  from  the
same  gene  as  a  result  of  alternative  splicing  are  called  protein  isoforms;;  see  Figure  1  In  fact,
about  95%  of  human  genes  are  commonly  spliced  in  more  than  one  way.  At  the  same  time,  when
alternative  splicing  goes  wrong,  it  can  create  the  same  negative  effects  caused  by  mutations,  and  it
has  been  blamed  for  a  number  of  genetic  disorders.

In  this  problem,  we  will  consider  a  simplified  model  of  alternative  splicing  in  which  any  of  a
collection  of  exons  can  be  chained  together  to  create  a  final  molecule  of  mRNA,  under  the
condition  that  we  use  a  minimum  number  of  exons  ( )  whose  order  is  fixed.  Because  the  exons
are  not  allowed  to  move  around,  we  need  only  select  a  subset  of  at  least    of  our  exons  to  chain
into  an  mRNA.

The  implied  computational  question  is  to  count  the  total  number  of  such  subsets,  which  will  provide
us  with  the  total  possible  number  of  alternatively  spliced  isoforms  for  this  model.

Problem

In  “Counting  Subsets”,  we  saw  that  the  total  number  of  subsets  of  a  set    containing    elements  is
equal  to   .

However,  if  we  intend  to  count  the  total  number  of  subsets  of    having  a  fixed  size   ,  then  we  use  the
combination  statistic   ,  also  written   .

Given:  Positive  integers    and    with   .

Return:  The  sum  of  combinations    for  all    satisfying   ,  modulo  1,000,000.  In
shorthand,   .

Sample  Dataset

6  3

Sample  Output
 42

Problem  52
Edit  Distance

Point  Mutations  Include  Insertions  and  Deletions

In  “Counting  Point  Mutations”,  we  saw  that  Hamming  distance  gave  us  a
preliminary  notion  of  the  evolutionary  distance  between  two  DNA  strings  by
counting  the  minimum  number  of  single  nucleotide  substitutions  that  could  have
occurred  on  the  evolutionary  path  between  the  two  strands.

However,  in  practice,  homologous  strands  of  DNA  or  protein  are  rarely  the  same  length  because
point  mutations  also  include  the  insertion  or  deletion  of  a  single  nucleotide  (and  single  amino  acids
can  be  inserted  or  deleted  from  peptides).  Thus,  we  need  to  incorporate  these  insertions  and
deletions  into  the  calculation  of  the  minimum  number  of  point  mutations  between  two  strings.  One
of  the  simplest  models  charges  a  unit  "cost"  to  any  single-­symbol  insertion/deletion,  then  (in
keeping  with  parsimony)  requests  the  minimum  cost  over  all  transformations  of  one  genetic  string
into  another  by  point  substitutions,  insertions,  and  deletions.

Problem

Given  two  strings    and    (of  possibly  different  lengths),  the  edit  distance    is  the  minimum
number  of  edit  operations  needed  to  transform    into   ,  where  an  edit  operation  is  defined  as  the
substitution,  insertion,  or  deletion  of  a  single  symbol.

The  latter  two  operations  incorporate  the  case  in  which  a  contiguous  interval  is  inserted  into  or  deleted
from  a  string;;  such  an  interval  is  called  a  gap.  For  the  purposes  of  this  problem,  the  insertion  or
deletion  of  a  gap  of  length    still  counts  as    distinct  edit  operations.

Given:  Two  protein  strings    and    in  FASTA  format  (each  of  length  at  most  1000  aa).
Return:  The  edit  distance   .

Sample  Dataset

>Rosalind_39
PLEASANTLY
>Rosalind_11
MEANLY

Sample  Output

5
Problem  53
Expected  Number  of  Restriction  Sites

A  Shot  in  the  Dark

In  “Locating  Restriction  Sites”,  we  first  familiarized  ourselves  with  restriction

enzymes.  Recall  that  these  enzymes  are  used  by  bacteria  to  cut  through  both
strands  of  viral  DNA,  thus  disarming  the  virus:  the  viral  DNA  locations  where  these
cuts  are  made  are  known  as  restriction  sites.  Recall  also  that  every  restriction
enzyme  is  preprogrammed  with  a  reverse  palindromic  interval  of  DNA  to  which  it  will  bind  and  cut,
called  a  recognition  sequence.  These  even  length  intervals  are  usually  either  4  or  6  base  pairs
long,  although  longer  ones  do  exist;;  rare-­cutter  enzymes  have  recognition  sequences  of  8  or  more
base  pairs.

In  this  problem,  we  will  ask  a  simple  question:  how  does  the  bacterium  "know"  that  it  will  probably
succeed  in  finding  a  restriction  site  within  the  virus's  DNA?  The  answer  is  that  the  short  length  of
recognition  sequences  guarantees  a  large  number  of  matches  occurring  randomly.

Intuitively,  we  would  expect  for  a  recognition  sequence  of  length  6  to  occur  on  average  once  every  
 base  pairs.  Note  that  this  fact  does  not  imply  that  the  associated  restriction  enzyme
will  cut  the  viral  DNA  every  4,096  bp;;  it  may  find  two  restriction  sites  close  together,  then  not  find  a
restriction  site  for  many  thousand  nucleotides.

In  this  problem,  we  will  generalize  the  problem  of  finding  an  average  number  of  restriction  sites  to
take  into  account  the  GC-­content  of  the  underlying  string  being  analyzed.

Problem

Say  that  you  place  a  number  of  bets  on  your  favorite  sports  teams.  If  their  chances  of  winning  are  0.3,
0.8,  and  0.6,  then  you  should  expect  on  average  to  win  0.3  +  0.8  +  0.6  =  1.7  of  your  bets  (of  course,
you  can  never  win  exactly  1.7!)

More  generally,  if  we  have  a  collection  of  events   ,  then  the  expected  number  of
events  occurring  is    (consult  the  note  following  the  problem  for  a
precise  explanation  of  this  fact).  In  this  problem,  we  extend  the  idea  of  finding  an  expected  number  of
events  to  finding  the  expected  number  of  times  that  a  given  string  occurs  as  a  substring  of  a  random
string.

Given:  A  positive  integer    ( ),  a  DNA  string    of  even  length  at  most  10,  and  an
array    of  length  at  most  20,  containing  numbers  between  0  and  1.

Return:  An  array    having  the  same  length  as    in  which    represents  the  expected  number  of
times  that    will  appear  as  a  substring  of  a  random  DNA  string    of  length   ,  where    is  formed  with
GC-­content    (see  “Introduction  to  Random  Strings”).

Sample  Dataset

10
AG
 0.25  0.5  0.75

Sample  Output

0.422  0.563  0.422

The  Mathematical  Details

In  this  problem,  we  are  speaking  of  an  expected  number  of  events;;  how  can  we  tie  this  into  the
definition  of  expected  value  that  we  already  have  from  “Calculating  Expected  Offspring”?

The  answer  relies  on  a  slick  mathematical  trick.  For  any  event   ,  we  can  form  a  random  variable
for   ,  called  an  indicator  random  variable   .  For  an  outcome   ,    when    belongs
to    and    when    belongs  to   .

For  an  indicator  random  variable   ,  verify  that   .

You  should  also  verify  from  our  original  formula  for  expected  value  that  for  any  two  random  variables
 and   ,    is  equal  to   .  As  a  result,  the  expected  number  of  events  
 occurring,  or   ,  reduces  to  
.

Problem  54
Motzkin  Numbers  and  RNA  Secondary  Structures

Dull,  Unhappy  People

In  “Catalan  Numbers  and  RNA  Secondary  Structures”,  we  talked  about  counting
the  number  of  ways  for  an  even  number  of  people  to  shake  hands  at  a  party
without  crossing  hands.  However,  in  the  real  world,  parties  only  contain  an  even
number  of  people  about  40%  of  the  time,  and  mathematicians  don't  enjoy
socializing.  So  we  should  instead  count  the  total  number  of  ways  for  some  of  the  people  at  the
party  to  shake  hands  without  crossing.

In  the  biological  world,  people  are  far  more  social,  but  not  every  nucleotide  in  a  strand  of  RNA  winds
up  base  pairing  with  another  nucleotide  during  RNA  folding.  As  a  result,  we  want  to  loosen  this
assumption  and  count  the  total  number  of  different  secondary  structures  of  an  RNA  strand  whose
base  pairs  don't  overlap  (i.e.,  we  still  forbid  pseudoknots  in  the  strand).

Problem

Similarly  to  our  definition  of  the  Catalan  numbers,  the   -­th  Motzkin  number    counts  the  number
of  ways  to  form  a  (not  necessarily  perfect)  noncrossing  matching  in  the  complete  graph  
containing    nodes.  For  example,  Figure  1  demonstrates  that   .  Note  in  this  figure  that
technically,  the  "trivial"  matching  that  contains  no  edges  at  all  is  considered  to  be  a  matching,
 because  it  satisfies  the  defining  condition  that  no  two  edges
are  incident  to  the  same  node.

How  should  we  compute  the  Motzkin  numbers?  As  with

Catalan  numbers,  we  will  take   .  To  calculate  
 in  general,  assume  that  the  nodes  of    are  labeled
around  the  outside  of  a  circle  with  the  integers  between  1  and   Figure  1.  The  21  distinct  ways  to
form  a  noncrossing  matching  on  5
,  and  consider  node  1,  which  may  or  may  not  be  involved  in labeled  nodes.
a  matching.  If  node  1  is  not  involved  in  a  matching,  then  there
are    ways  of  matching  the  remaining    nodes.  If
node  1  is  involved  in  a  matching,  then  say  it  is  matched  to
node   :  this  leaves    nodes  on  one  side  of  edge  
and    nodes  on  the  other  side;;  as  with  the  Catalan
numbers,  no  edge  can  connect  the  two  sides,  which  gives  us  
 ways  of  matching  the  remaining  edges.
Allowing    to  vary  between    and    yields  the  following Figure  2.  Two  possible  noncrossing
matchings  of  basepair  edges  in  the
recurrence  relation  for  the  Motzkin  numbers:   bonding  graph  corresponding  to
. RNA  string  UAGCGUGAUCAC.

To  count  all  possible  secondary  structures  of  a  given  RNA

string  that  do  not  contain  pseudoknots,  we  need  to  modify  the  Motzkin  recurrence  so  that  it  counts
only  matchings  of  basepair  edges  in  the  bonding  graph  corresponding  to  the  RNA  string;;  see  Figure  2

Given:  An  RNA  string    of  length  at  most  300  bp.
Return:  The  total  number  of  noncrossing  matchings  of  basepair  edges  in  the  bonding  graph  of   ,
modulo  1,000,000.

Sample  Dataset

>Rosalind_57
AUAU

Sample  Output

Problem  55
Distances  in  Trees

Paths  in  Trees

For  any  two  nodes  of  a  tree,  a  unique  path  connects  the  nodes;;  more  specifically,
there  is  a  unique  path  connecting  any  pair  of  leaves.  Why  must  this  be  the  case?
If  more  than  one  path  connected  two  nodes,  then  they  would  necessarily  form  a
cycle,  which  would  violate  the  definition  of  tree.
 The  uniqueness  of  paths  connecting  nodes  in  a  tree  is  helpful  in  phylogenetic  analysis  because  a
rudimentary  measure  of  the  separation  between  two  taxa  is  the  distance  between  them  in  the  tree,
which  is  equal  to  the  number  of  edges  on  the  unique  path  connecting  the  two  leaves  corresponding
to  the  taxa.

Problem

Newick  format  is  a  way  of  representing  trees  even  more

concisely  than  using  an  adjacency  list,  especially  when
dealing  with  trees  whose  internal  nodes  have  not  been  labeled.

First,  consider  the  case  of  a  rooted  tree   .  A  collection  of

leaves    of    are  neighbors  if  they  are  all
adjacent  to  some  internal  node   .  Newick  format  for    is
obtained  by  iterating  the  following  key  step:  delete  all  the
edges    from    and  label    with   .
This  process  is  repeated  all  the  way  to  the  root,  at  which  point
a  semicolon  signals  the  end  of  the  tree.

A  number  of  variations  of  Newick  format  exist.  First,  if  a  node
is  not  labeled  in   ,  then  we  simply  leave  blank  the  space Figure  1.  This  tree  can  be
represented  in  Newick  format  in  a
occupied  by  the  node.  In  the  key  step,  we  can  write   number  of  ways,  including  (C,  D,  (A,
 in  place  of    if  the    are B));;  (A,  (D,  C),  B);;  and  (((A,  B),  C))D;;.
labeled;;  if  none  of  the  nodes  are  labeled,  we  can  write  
.

A  second  variation  of  Newick  format  occurs  when    is  unrooted,  in  which  case  we  simply  select  any
internal  node  to  serve  as  the  root  of   .  A  particularly  peculiar  case  of  Newick  format  arises  when  we
choose  a  leaf  to  serve  as  the  root.

Note  that  there  will  be  a  large  number  of  different  ways  to  represent    in  Newick  format;;  see  Figure  1.

Given:  A  collection  of    trees  ( )  in  Newick  format,  with  each  tree  containing  at  most  200
nodes;;  each  tree    is  followed  by  a  pair  of  nodes    and    in   .

Return:  A  collection  of    positive  integers,  for  which  the   th  integer  represents  the  distance
between    and    in   .

Sample  Dataset

(cat)dog;
dog  cat

(dog,cat);
dog  cat

Sample  Output

1  2
Problem  56
Interleaving  Two  Motifs

Two  Motifs,  One  Gene

Recall  that  in  “Finding  a  Shared  Spliced  Motif”,  we  found  the  longest  motif  that
could  have  been  shared  by  two  genetic  strings,  allowing  for  the  motif  to  be  split
onto  multiple  exons  in  the  process.  As  a  result,  we  needed  to  find  a  longest
common  subsequence  of  the  two  strings  (which  extended  the  problem  of  finding  a
longest  common  substring  from  “Finding  a  Shared  Motif”).

In  this  problem,  we  consider  an  inverse  problem  of  sorts  in  which  we  are  given  two  different  motifs
and  we  wish  to  interleave  them  together  in  such  a  way  as  to  produce  a  shortest  possible  genetic
string  in  which  both  motifs  occur  as  subsequences.

Problem

A  string    is  a  supersequence  of  another  string    if    contains    as  a  subsequence.

A  common  supersequence  of  strings    and    is  a  string  that  serves  as  a  supersequence  of  both  
and   .  For  example,  "GACCTAGGAACTC"  serves  as  a  common  supersequence  of  "ACGTC"  and
"ATAT".  A  shortest  common  supersequence  of    and    is  a  supersequence  for  which  there  does  not
exist  a  shorter  common  supersequence.  Continuing  our  example,  "ACGTACT"  is  a  shortest  common
supersequence  of  "ACGTC"  and  "ATAT".

Given:  Two  DNA  strings    and   .

Return:  A  shortest  common  supersequence  of    and   .  If  multiple  solutions  exist,  you  may  output
any  one.

Sample  Dataset

ATCTGAT
TGCATA

Sample  Output

ATGCATGAT

Problem  57
Sorting  by  Reversals
 Reconstructing  Evolutionary  Histories

When  we  calculate  the  Hamming  distance  between  two  genetic  strings,  we  can
easily  infer  a  collection  of  point  mutations  that  occurred  on  the  evolutionary  path
between  the  two  strings  by  simply  examining  the  mismatched  symbols.  However,
when  calculating  the  reversal  distance  (see  “Reversal  Distance”),  we  only  have  the
minimum  number  of  reversals  separating  two  permutations.

The  computational  problem  of  sorting  by  reversals  demands  instead  that  we  provide  a  minimum
list  of  reversals  transforming  one  permutation  into  another.  As  a  result,  sorting  by  reversals
subsumes  the  calculation  of  reversal  distance:  once  we  have  a  minimum  collection  of  reversals,  the
reversal  distance  follows  immediately.

Problem

A  reversal  of  a  permutation  can  be  encoded  by  the  two  indices  at  the  endpoints  of  the  interval  that  it
inverts;;  for  example,  the  reversal  that  transforms    into    is  encoded  by  
.

A  collection  of  reversals  sorts    into    if  the  collection  contains    reversals,  which  when
successively  applied  to    yield   .

Given:  Two  permutations    and   ,  each  of  length  10.

Return:  The  reversal  distance   ,  followed  by  a  collection  of  reversals  sorting    into   .  If
multiple  collections  of  such  reversals  exist,  you  may  return  any  one.

Sample  Dataset

1  2  3  4  5  6  7  8  9  10
1  8  9  3  2  7  6  5  4  10

Sample  Output

2
4  9
2  5

Problem  58
Inferring  Protein  from  Spectrum

Introduction  to  Mass  Spectrometry

 In  “Calculating  Protein  Mass”,  we  briefly  mentioned  an  analytic  chemical  method
called  mass  spectrometry,  which  aims  to  measure  the  mass-­to-­charge  ratio  of  a
particle  or  a  molecule.  In  a  mass  spectrometer,  a  sample  is  vaporized  (turned  into
gas),  and  then  particles  from  the  sample  are  ionized.  The  resulting  ions  are
placed  into  an  electromagnetic  field,  which  separates  them  based  on  their  charge
and  mass.  The  output  of  the  mass  spectrometer  is  a  mass  spectrum,  or  a  plot  of
ions'  possible  mass-­to-­charge  ratio  values  with  the  intensity  (actual  observed
frequency)  of  ions  having  these  mass-­to-­charge  values.

For  the  moment,  we  will  ignore  charge  and  consider  a  list  of  the  ions'  monoisotopic  masses  as  a
simplified  spectrum.  Researchers  do  not  possess  cheap  technology  to  go  in  and  examine  a
protein  one  amino  acid  at  a  time  (molecules  are  too  submicroscopic).  Instead,  to  determine  a
protein's  structure,  we  will  split  several  copies  of  the  protein  into  smaller  pieces,  then  weigh  the
resulting  fragments.  To  do  this,  we  assume  that  each  cut  (breakage  point)  occurs  between  two
amino  acids  and  that  we  can  measure  the  mass  of  the  resulting  pieces  for  all  possible  cuts.

For  example,  the  (unknown)  protein  "PRTEIN"  can  be  cut  in  five  possible  ways:  "P"  and  "RTEIN";;
"PR"  and  "TEIN";;  "PRT"  and  "EIN";;  "PRTE"  and  "IN";;  "PRTEI"  and  "N".  We  then  can  measure  the
masses  of  all  fragments,  including  the  entire  string.  The  "left"  end  of  a  protein  is  called  its  N-­
terminus,  and  the  ions  corresponding  to  the  protein  string's  prefixes  (P,  PR,  PRT,  PRTE,  PRTEI)
are  called  b-­ions.  The  "right"  end  of  the  protein  is  called  its  C-­terminus,  and  the  ions
corresponding  to  the  string's  suffixes  (N,  IN,  EIN,  TEIN,  RTEIN)  are  called  y-­ions.  The  difference  in
the  masses  of  two  adjacent  b-­ions  (or  y-­ions)  gives  the  mass  of  one  amino  acid  residue;;  for
example,  the  difference  between  the  masses  of  "PRT"  and  "PR"  must  be  the  mass  of  "T."  By
extension,  knowing  the  masses  of  every  b-­ion  of  a  protein  allows  us  to  deduce  the  protein's
identity.

Problem

The  prefix  spectrum  of  a  weighted  string  is  the  collection  of  all  its  prefix  weights.

Given:  A  list    of    ( )  positive  real  numbers.

Return:  A  protein  string  of  length    whose  prefix  spectrum  is  equal  to    (if  multiple  solutions
exist,  you  may  output  any  one  of  them).  Consult  the  monoisotopic  mass  table.

Sample  Dataset

3524.8542
3710.9335
3841.974
3970.0326
4057.0646

Sample  Output

WMQS

Problem  59
Introduction  to  Pattern  Matching

If  At  First  You  Don't  Succeed...

We  introduced  the  problem  of  finding  a  motif  in  a  genetic  string  in  “Finding  a  Motif
in  DNA”.  More  commonly,  we  will  have  a  collection  of  motifs  that  we  may  wish  to
find  in  a  larger  string,  for  example  when  searching  a  genome  for  a  collection  of
known  genes.

This  application  sets  up  the  algorithmic  problem  of  pattern  matching,  in  which  we  are  searching  a
large  string  (called  a  text)  for  instances  of  a  collection  of  smaller  strings,  called  patterns.  For  the
moment,  we  will  focus  on  requiring  that  all  matches  should  be  exact.

The  most  obvious  method  for  finding  exact  patterns  in  a  text  is  to  simply  apply  a  simple  "sliding
window"  algorithm  for  each  pattern.  However,  this  method  is  time-­consuming  if  we  have  a  large
number  of  patterns  to  consider  (which  will  often  be  the  case  when  dealing  with  a  database  of
genes).  It  would  be  better  if  instead  of  traversing  the  genome  for  every  pattern,  we  could  somehow
only  traverse  it  once.  To  this  end,  we  will  need  a  data  structure  that  can  efficiently  organize  a
collection  of  patterns.

Problem

Given  a  collection  of  strings,  their  trie  (often  pronounced  "try"

to  avoid  ambiguity  with  the  general  term  tree)  is  a  rooted  tree
formed  as  follows.  For  every  unique  first  symbol  in  the  strings,
an  edge  is  formed  connecting  the  root  to  a  new  vertex.  This
symbol  is  then  used  to  label  the  edge.

We  may  then  iterate  the  process  by  moving  down  one  level  as
follows.  Say  that  an  edge  connecting  the  root  to  a  node    is
labeled  with  'A';;  then  we  delete  the  first  symbol  from  every
string  in  the  collection  beginning  with  'A'  and  then  treat    as
our  root.  We  apply  this  process  to  all  nodes  that  are  adjacent
to  the  root,  and  then  we  move  down  another  level  and
continue.  See  Figure  1  for  an  example  of  a  trie. Figure  1.  The  trie  corresponding  to
the  strings  'apple',  'apropos',
As  a  result  of  this  method  of  construction,  the  symbols  along 'banana',  'bandana',  and  'orange'.
the  edges  of  any  path  in  the  trie  from  the  root  to  a  leaf  will Each  path  from  root  to  leaf  encodes
spell  out  a  unique  string  from  the  collection,  as  long  as  no one  of  these  strings.
string  is  a  prefix  of  another  in  the  collection  (this  would  cause
the  first  string  to  be  encoded  as  a  path  terminating  at  an  internal  node).

Given:  A  list  of  at  most  100  DNA  strings  of  length  at  most  100  bp,  none  of  which  is  a  prefix  of
another.

Return:  The  adjacency  list  corresponding  to  the  trie    for  these  patterns,  in  the  following  format.  If  
 has    nodes,  first  label  the  root  with  1  and  then  label  the  remaining  nodes  with  the  integers  2
through    in  any  order  you  like.  Each  edge  of  the  adjacency  list  of    will  be  encoded  by  a  triple
containing  the  integer  representing  the  edge's  parent  node,  followed  by  the  integer  representing  the
edge's  child  node,  and  finally  the  symbol  labeling  the  edge.

Sample  Dataset
 ATAGA
ATC
GAT

Sample  Output

1  2  A
2  3  T
3  4  A
4  5  G
5  6  A
3  7  C
1  8  G
8  9  A
9  10  T

Problem  60
Comparing  Spectra  with  the  Spectral  Convolution

Comparing  Spectra

Suppose  you  have  two  mass  spectra,  and  you  want  to  check  if  they  both  were
obtained  from  the  same  protein;;  you  will  need  some  notion  of  spectra  similarity.
The  simplest  possible  metric  would  be  to  count  the  number  of  peaks  in  the  mass
spectrum  that  the  spectra  share,  called  the  shared  peaks  count;;  its  analogue  for
simplified  spectra  is  the  number  of  masses  that  the  two  spectra  have  in  common.

The  shared  peaks  count  can  be  useful  in  the  simplest  cases,  but  it  does  not  help  us  if,  for
example,  one  spectrum  corresponds  to  a  peptide  contained  inside  of  another  peptide  from  which
the  second  spectrum  was  obtained.  In  this  case,  the  two  spectra  are  very  similar,  but  the  shared
peaks  count  will  be  very  small.  However,  if  we  shift  one  spectrum  to  the  right  or  left,  then  shared
peaks  will  align.  In  the  case  of  simplified  spectra,  this  means  that  there  is  some  shift  value    such
that  adding    to  the  weight  of  every  element  in  one  spectrum  should  create  a  large  number  of
matches  in  the  other  spectrum.

Problem

A  multiset  is  a  generalization  of  the  notion  of  set  to  include  a  collection  of  objects  in  which  each
object  may  occur  more  than  once  (the  order  in  which  objects  are  given  is  still  unimportant).  For  a
multiset   ,  the  multiplicity  of  an  element    is  the  number  of  times  that    occurs  in  the  set;;  this
multiplicity  is  denoted   .  Note  that  every  set  is  included  in  the  definition  of  multiset.

The  Minkowski  sum  of  multisets    and    containing  real  numbers  is  the  new  multiset  
formed  by  taking  all  possible  sums    of  an  element    from    and  an  element    from   .
The  Minkowski  sum  could  be  defined  more  concisely  as  
 ,  The  Minkowski  difference    is  defined
analogously  by  taking  all  possible  differences   .

If    and    represent  simplified  spectra  taken  from  two  peptides,  then    is  called  the
spectral  convolution  of    and   .  In  this  notation,  the  shared  peaks  count  is  represented  by  
,  and  the  value  of    for  which    has  the  maximal  value  is  the  shift  value
maximizing  the  number  of  shared  masses  of    and   .

Given:  Two  multisets  of  positive  real  numbers    and   .  The  size  of  each  multiset  is  at  most
200.

Return:  The  largest  multiplicity  of   ,  as  well  as  the  absolute  value  of  the  number  
maximizing    (you  may  return  any  such  value  if  multiple  solutions  exist).

Sample  Dataset

186.07931  287.12699  548.20532  580.18077  681.22845  706.27446  782.27613  9

68.35544  968.35544
101.04768  158.06914  202.09536  318.09979  419.14747  463.17369

Sample  Output

3
85.03163

Note

Observe  that    is  equivalent  to   ,  but  it  is  not  usually  the  case  that    is  the
same  as   ;;  in  this  case,  one  multiset  can  be  obtained  from  the  other  by  negating  every
element.

Problem  61
Creating  a  Character  Table

Introduction  to  Character-­Based  Phylogeny

Before  the  modern  genetics  revolution,  phylogenies  were  constructed  from

physical  characters  resulting  from  direct  structural  comparison  of  taxa.  A  great
deal  of  analysis  relied  on  the  fossil  record,  as  fossils  provided  the  only  concrete
framework  for  studying  the  appearance  of  extinct  species  and  for  inferring  how
they  could  have  evolved  into  present-­day  organisms.

A  classic  case  illustrating  the  utility  of  the  fossil  record  is  the  case  of  dinosaur  pelvic  bones.  In
1887,  Harry  Seeley  proposed  a  new  classification  of  dinosaurs  into  two  orders,  Saurischia  and
Ornithischia:  the  former  possessed  hip  bones  shaped  like  those  found  in  reptiles,  whereas  the
latter  had  a  much  different  hip  shape  that  resembled  birds.  Seeley's  pelvic  classification  has
 survived  to  the  present  day  as  the  principal  division  of  dinosaurs.

The  key  point  is  that  hip  bone  shape  is  a  physical  character  that  separates  all  dinosaurs  into  two
different  groups.  Our  hope  is  to  construct  a  phylogeny  solely  from  a  collection  of  characters.
Throughout  character-­based  phylogeny,  our  two-­part  assumption  is  that  all  taxa  possessing  a
character  must  have  evolved  from  a  single  ancestor  that  introduced  this  character,  and  conversely,
any  taxon  not  possessing  the  character  cannot  be  descended  from  this  ancestor.

Problem

Given  a  collection  of    taxa,  any  subset    of  these  taxa  can  be  seen  as  encoding  a  character  that
divides  the  taxa  into  the  sets    and   ;;  we  can  represent  the  character  by   ,  which  is  called  a
split.  Alternately,  the  character  can  be  represented  by  a  character  array    of  length    for  which  
 if  the   th  taxon  belongs  to    and    if  the   th  taxon  belongs  to    (recall  the
"ON"/"OFF"  analogy  from  “Counting  Subsets”).

At  the  same  time,  observe  that  the  removal  of  an  edge  from  an  unrooted  binary  tree  produces  two
separate  trees,  each  one  containing  a  subset  of  the  original  taxa.  So  each  edge  may  also  be  encoded
by  a  split   .

A  trivial  character  isolates  a  single  taxon  into  a  group  of  its  own.  The  corresponding  split  
must  be  such  that    or    contains  only  one  element;;  the  edge  encoded  by  this  split  must  be
incident  to  a  leaf  of  the  unrooted  binary  tree,  and  the  array  for  the  character  contains  exactly  one  0  or
exactly  one  1.  Trivial  characters  are  of  no  phylogenetic  interest  because  they  fail  to  provide  us  with
information  regarding  the  relationships  of  taxa  to  each  other.  All  other  characters  are  called  nontrivial
characters  (and  the  associated  splits  are  called  nontrivial  splits).

A  character  table  is  a  matrix    in  which  each  row  represents  the  array  notation  for  a  nontrivial
character.  That  is,  entry    denotes  the  "ON"/"OFF"  position  of  the   th  character  with  respect  to  the  
th  taxon.

Given:  An  unrooted  binary  tree    in  Newick  format  for  at  most  200  species  taxa.
Return:  A  character  table  having  the  same  splits  as  the  edge  splits  of   .  The  columns  of  the
character  table  should  encode  the  taxa  ordered  lexicographically;;  the  rows  of  the  character  table  may
be  given  in  any  order.  Also,  for  any  given  character,  the  particular  subset  of  taxa  to  which  1s  are
assigned  is  arbitrary.

Sample  Dataset

(dog,((elephant,mouse),robot),cat);

Sample  Output

00110
00111

Problem  62
Constructing  a  De  Bruijn  Graph

Wading  Through  the  Reads

Because  we  use  multiple  copies  of  the  genome  to  generate  and  identify  reads  for
the  purposes  of  fragment  assembly,  the  total  length  of  all  reads  will  be  much
longer  than  the  genome  itself.  This  begs  the  definition  of  read  coverage  as  the
average  number  of  times  that  each  nucleotide  from  the  genome  appears  in  the
reads.  In  other  words,  if  the  total  length  of  our  reads  is  30  billion  bp  for  a  3  billion  bp  genome,  then
we  have  10x  read  coverage.

To  handle  such  a  large  number  of   -­mers  for  the  purposes  of  sequencing  the  genome,  we  need  an
efficient  and  simple  structure.

Problem

Consider  a  set    of   -­mers  of  some  unknown  DNA  string.  Let    denote  the  set  containing
all  reverse  complements  of  the  elements  of   .  (recall  from  “Counting  Subsets”  that  sets  are  not
allowed  to  contain  duplicate  elements).

The  de  Bruijn  graph    of  order    corresponding  to    is  a  digraph  defined  in  the  following
way:

Nodes  of    correspond  to  all   -­mers  that  are  present  as  a  substring  of  a   -­mer  from  
.
Edges  of    are  encoded  by  the   -­mers  of    in  the  following  way:  for  each  
-­mer    in   ,  form  a  directed  edge  ( ,   ).

Given:  A  collection  of  up  to  1000  DNA  strings  of  equal  length  (not  exceeding  50  bp)  corresponding
to  a  set    of   -­mers.

Return:  The  adjacency  list  corresponding  to  the  de  Bruijn  graph  corresponding  to   .

Sample  Dataset

TGAT
CATG
TCAT
ATGC
CATC
CATC

Sample  Output

(ATC,  TCA)
(ATG,  TGA)
(ATG,  TGC)
(CAT,  ATC)
(CAT,  ATG)
(GAT,  ATG)
 (GCA,  CAT)
(TCA,  CAT)
(TGA,  GAT)

Problem  63
Edit  Distance  Alignment

Reconstructing  Edit  Distance

In  “Counting  Point  Mutations”,  the  calculation  of  Hamming  distance  gave  us  a
clear  way  to  model  the  sequence  of  point  mutations  transforming  one  genetic
string  into  another.  By  simply  writing  one  string  directly  over  the  other,  we  could
count  each  mismatched  symbol  as  a  substitution.

However,  in  the  calculation  of  edit  distance  (see  “Edit  Distance”),  the  two  strings  can  have  different
lengths;;  thus,  simply  superimposing  one  string  over  the  other  does  us  no  good  when  it  comes  to
visualizing  a  sequence  of  edit  operations  transforming  one  string  into  the  other.  To  remedy  this,  we
will  introduce  a  new  symbol  to  serve  as  a  placeholder  representing  an  inserted  or  deleted  symbol;;
furthermore,  this  placeholder  will  allow  us  to  align  two  strings  of  differing  lengths.

Problem

An  alignment  of  two  strings    and    is  defined  by  two  strings    and    satisfying  the  following  three
conditions:  1.    and    must  be  formed  from  adding  gap  symbols  "-­"  to  each  of    and   ,  respectively;;
as  a  result,    and    will  form  subsequences  of    and   .  2.    and    must  have  the  same  length.  3.
Two  gap  symbols  may  not  be  aligned;;  that  is,  if    is  a  gap  symbol,  then    cannot  be  a  gap
symbol,  and  vice-­versa.

We  say  that    and    augment    and   .  Writing    directly  over    so  that  symbols  are  aligned
provides  us  with  a  scenario  for  transforming    into   .  Mismatched  symbols  from    and    correspond  to
symbol  substitutions;;  a  gap  symbol    aligned  with  a  non-­gap  symbol    implies  the  insertion  of
this  symbol  into   ;;  a  gap  symbol    aligned  with  a  non-­gap  symbol    implies  the  deletion  of  this
symbol  from   .

Thus,  an  alignment  represents  a  transformation  of    into    via  edit  operations.  We  define  the
corresponding  edit  alignment  score  of    and    as    (Hamming  distance  is  used  because
the  gap  symbol  has  been  introduced  for  insertions  and  deletions).  It  follows  that  
,  where  the  minimum  is  taken  over  all  alignments  of    and   .  We  call
such  a  minimum  score  alignment  an  optimal  alignment  (with  respect  to  edit  distance).

Given:  Two  protein  strings    and    in  FASTA  format  (with  each  string  having  length  at  most  1000
aa).

Return:  The  edit  distance    followed  by  two  augmented  strings    and    representing  an
optimal  alignment  of    and   .

Sample  Dataset
 >Rosalind_43
PRETTY
>Rosalind_97
PRTTEIN

Sample  Output

4
PRETTY-­‐-­‐
PR-­‐TTEIN

Problem  64
Inferring  Peptide  from  Full  Spectrum

Ions  Galore

In  “Inferring  Protein  from  Spectrum”,  we  inferred  a  protein  string  from  a  list  of  b-­
ions.  In  practice,  biologists  have  no  way  of  distinguishing  between  b-­ions  and  y-­
ions  in  the  simplified  spectrum  of  a  peptide.  However,  we  will  often  possess  a  pair
of  masses  in  the  spectrum  corresponding  to  a  single  cut.  The  two  corresponding
ions  complement  each  other:  for  example,  mass("PR")  +  mass("TEIN")  =  mass("PRTEIN").  As  a
result,  we  can  easily  infer  the  mass  of  a  b-­ion  from  its  complementary  y-­ion  and  vice  versa,  as  long
as  we  already  know  the  parent  mass,  i.e.,  the  mass  of  the  entire  peptide.

The  theoretical  simplified  spectrum  for  a  protein    of  length    is  constructed  as  follows:  form  all
possible  cuts,  then  compute  the  mass  of  the  b-­ion  and  the  y-­ion  at  each  cut.  Duplicate  masses  are
allowed.  You  might  guess  how  we  could  modify  “Inferring  Protein  from  Spectrum”  to  infer  a  peptide
from  its  theoretical  simplified  spectrum;;  here  we  consider  a  slightly  modified  form  of  this  problem  in
which  we  attempt  to  identify  the  interior  region  of  a  peptide  given  only  b-­ions  and  y-­ions  that  are  cut
within  this  region.  As  a  result,  we  will  have  constant  masses  at  the  beginning  and  end  of  the
peptide  that  will  be  present  in  the  mass  of  every  b-­ion  and  y-­ion,  respectively.

Problem

Say  that  we  have  a  string    containing    as  an  internal  substring,  so  that  there  exist  nonempty
substrings    and    of    such  that    can  be  written  as   .  A  t-­prefix  contains  all  of    and  none
of   ;;  likewise,  a  t-­suffix  contains  all  of    and  none  of   .

Given:  A  list    containing    positive  real  numbers  ( ).  The  first  number  in    is  the
parent  mass  of  a  peptide   ,  and  all  other  numbers  represent  the  masses  of  some  b-­ions  and  y-­ions  of
 (in  no  particular  order).  You  may  assume  that  if  the  mass  of  a  b-­ion  is  present,  then  so  is  that  of  its
complementary  y-­ion,  and  vice-­versa.

Return:  A  protein  string    of  length    for  which  there  exist  two  positive  real  numbers    and  
such  that  for  every  prefix    and  suffix    of   ,  each  of    and    is  equal  to  an
 element  of   .  (In  other  words,  there  exists  a  protein  string  whose   -­prefix  and   -­suffix  weights
correspond  to  the  non-­parent  mass  values  of   .)  If  multiple  solutions  exist,  you  may  output  any  one.

Sample  Dataset

1988.21104821
610.391039105
738.485999105
766.492149105
863.544909105
867.528589105
992.587499105
995.623549105
1120.6824591
1124.6661391
1221.7188991
1249.7250491
1377.8200091

Sample  Output

KEKEP

Problem  65
Independent  Segregation  of  Chromosomes

Mendel's  Work  Examined

Mendel's  laws  of  heredity  were  initially  ignored,  as  only  11  papers  have  been
found  that  cite  his  paper  between  its  publication  in  1865  and  1900.  One  reason  for
Mendel's  lack  of  popularity  is  that  information  did  not  move  quite  so  readily  as  in
the  modern  age;;  perhaps  another  reason  is  that  as  a  friar  in  an  Austrian  abbey,
Mendel  was  already  isolated  from  Europe's  university  community.

It  is  fair  to  say  that  no  one  who  did  initially  read  Mendel's  work  fully  believed  that  traits  for  more
complex  organisms,  like  humans,  could  be  broken  down  into  discrete  units  of  heredity  (i.e.,
Mendel's  factors).  This  skepticism  was  well-­founded  in  empirical  studies  of  inheritance,  which
indicated  a  far  more  complex  picture  of  heredity  than  Mendel's  theory  dictated.  The  friar  himself
admitted  that  representing  every  trait  with  a  single  factor  was  overly  simplistic,  and  so  he  proposed
that  some  traits  are  polymorphic,  or  encoded  by  multiple  different  factors.

Yet  any  hereditary  model  would  ultimately  be  lacking  without  an  understanding  of  how  traits  are
physically  passed  from  organisms  to  their  offspring.  This  physical  mechanism  was  facilitated  by
Walther  Flemming's  1879  discovery  of  chromosomes  in  salamander  eggs  during  cell  division,
followed  by  Theodor  Boveri's  observation  that  sea  urchin  embryos  with  chromatin  removed  failed  to
 develop  correctly  (implying  that  traits  must  somehow  be  encoded  on  chromosomes).  By  the  turn  of
the  20th  century,  Mendel's  work  had  been  rediscovered  by  Hugo  de  Vries  and  Carl  Correns,  but  it
was  still  unclear  how  Mendel's  hereditary  model  could  be  tied  to  chromosomes.

Fortunately,  Walter  Sutton  demonstrated  that  grasshopper  chromosomes  occur  in  matched  pairs
called  homologous  chromosomes,  or  homologs.  We  now  know  that  the  DNA  found  on
homologous  chromosomes  is  identical  except  for  minor  variations  attributable  to  SNPs  and  small
rearrangements,  which  are  typically  insertions  and  deletions.  Sutton  himself,  working  five  decades
before  Watson  &  Crick  and  possessing  no  real  understanding  of  DNA,  actually  surmised  that
variations  to  homologous  chromosomes  should  somehow  correspond  to  Mendel's  alleles.

Yet  it  still  remained  to  show  how  chromosomes  themselves  are  inherited.  Most  multicellular
organisms  are  diploid,  meaning  that  their  cells  possess  two  sets  of  chromosomes;;  humans  are
included  among  diploid  organisms,  having  23  homologous  chromosome  pairs.

Gametes  (i.e.,  sex  cells)  in  diploid  organisms  form  an  exception  and  are  haploid,  meaning  that
they  only  possess  one  chromosome  from  each  pair  of  homologs.  During  the  fusion  of  two  gametes
of  opposite  sex,  a  diploid  embryo  is  formed  by  simply  uniting  the  two  gametes'  halved  chromosome
sets.

Mendel's  first  law  can  now  be  explained  by  the  fact  that  during  the  meiosis  each  gamete  randomly
selects  one  of  the  two  available  alleles  of  the  particular  gene.

Mendel's  second  law  follows  from  the  fact  that  gametes  select  nonhomologous  chromosomes
independently  of  each  other;;  however,  this  law  will  hold  only  for  factors  encoded  on  nonhomologous
chromosomes,  which  leaves  open  the  inheritance  of  factors  encoded  on  homologous
chromosomes.

Problem

Consider  a  collection  of  coin  flips.  One  of  the  most  natural  questions  we  can  ask  is  if  we  flip  a  coin  92
times,  what  is  the  probability  of  obtaining  51  "heads",  vs.  27  "heads",  vs.  92  "heads"?

Each  coin  flip  can  be  modeled  by  a  uniform  random  variable  in  which  each  of  the  two  outcomes
("heads"  and  "tails")  has  probability  equal  to  1/2.  We  may  assume  that  these  random  variables  are
independent  (see  “Independent  Alleles”);;  in  layman's  terms,  the  outcomes  of  the  two  coin  flips  do  not
influence  each  other.

A  binomial  random  variable    takes  a  value  of    if    consecutive  "coin  flips"  result  in    total
"heads"  and    total  "tails."  We  write  that   .

Given:  A  positive  integer   .

Return:  An  array    of  length    in  which    represents  the  common  logarithm  of  the  probability
that  two  diploid  siblings  share  at  least    of  their    chromosomes  (we  do  not  consider  recombination
for  now).

Sample  Dataset

Sample  Output

0.000  -­‐0.004  -­‐0.024  -­‐0.082  -­‐0.206  -­‐0.424  -­‐0.765  -­‐1.262  -­‐1.969  -­‐3.010
Problem  66
Finding  Disjoint  Motifs  in  a  Gene

Disjoint  Motifs

In  this  problem,  we  will  consider  an  algorithmic  (but  not  particularly  practical)
variant  of  motif  finding  for  multiple  motifs.  Say  we  have  two  motifs  corresponding
to  the  coding  regions  of  genes,  and  we  want  to  know  whether  these  motifs  can  be
found  in  genes  occupying  the  same  region  of  the  genome.  To  prevent  exons  from
coinciding,  we  further  stipulate  that  the  two  motifs  are  nonoverlapping.

In  this  problem,  we  will  ask  whether  two  disjoint  motifs  can  be  located  in  a  given  string.  We
considered  a  similar  problem  in  “Interleaving  Two  Motifs”,  which  asked  us  to  find  a  shortest  possible
string  containing  two  motifs;;  however,  in  that  problem,  the  motifs  were  allowed  to  coincide.

Problem

Given  three  strings   ,   ,  and   ,  we  say  that    and    can  be  interwoven  into    if  there  is  some
substring  of    made  up  of    and    as  disjoint  subsequences.

For  example,  the  strings  " "  and  " "  can  be  interwoven  into  " ".
However,  they  cannot  be  interwoven  into  " "  because  of  the  appearance  of
the  four  'A's  in  the  middle  of  the  subsequences.  Similarly,  even  though  both  " "  is  a  shortest
common  supersequence  of    and   ,  it  is  not  possible  to  interweave  these  two  strings  into
" "  because  the  two  desired  subsequences  must  be  disjoint;;  see  “Interleaving  Two  Motifs”  for
details  on  finding  a  shortest  common  supersequence  of  two  strings.

Given:  A  text  DNA  string    of  length  at  most  10  kbp,  followed  by  a  collection  of    ( )  DNA
strings  of  length  at  most  10  bp  acting  as  patterns.

Return:  An    matrix    for  which    if  the   th  and   th  pattern  strings  can  be
interwoven  into    and    otherwise.

Sample  Dataset

GACCACGGTT
ACAG
GT
CCG

Sample  Output

0  0  1
0  1  0
1  0  0
 Citation

This  problem  follows  Jones  &  Pevzner,  An  Introduction  to  Bioinformatics  Algorithms,  Problem  6.31

Problem  67
Finding  the  Longest  Multiple  Repeat

Long  Repeats

We  saw  in  “Introduction  to  Pattern  Matching”  that  a  data  structure  commonly
used  to  encode  the  relationships  among  a  collection  of  strings  was  the  trie,  which
is  particularly  useful  when  the  strings  represent  a  collection  of  patterns  that  we
wish  to  match  to  a  larger  text.

The  trie  is  helpful  when  processing  multiple  strings  at  once,  but  when  we  want  to  analyze  a  single
string,  we  need  something  different.

In  this  problem,  we  will  use  a  new  data  structure  to  handle  the  problem  of  finding  long  repeats  in  the
genome.  Recall  from  “Finding  a  Motif  in  DNA”  that  cataloguing  these  repeats  is  a  problem  of  the
utmost  interest  to  molecular  biologists,  as  a  natural  correlation  exists  between  the  frequency  of  a
repeat  and  its  influence  on  RNA  transcription.  Our  aim  is  therefore  to  identify  long  repeats  that
occur  more  than  some  predetermined  number  of  times.

Problem

A  repeated  substring  of  a  string    of  length    is  simply  a

substring  that  appears  in  more  than  one  location  of   ;;  more
specifically,  a  k-­fold  substring  appears  in  at  least  k  distinct
locations.

The  suffix  tree  of   ,  denoted   ,  is  defined  as  follows:

Figure  1.  The  suffix  tree  for  s  =
 is  a  rooted  tree  having  exactly    leaves. GTCCGAAGCTCCGG.  Note  that  the
Every  edge  of    is  labeled  with  a  substring  of   , dollar  sign  has  been  appended  to  a
where    is  the  string  formed  by  adding  a  placeholder substring  of  the  tree  to  mark  the  end
of  s.  Every  path  from  the  root  to  a  leaf
symbol   $  to  the  end  of   . corresponds  to  a  unique  suffix  of
Every  internal  node  of    other  than  the  root  has  at GTCCGAAGCTCCGG,  and  each  leaf
is  labeled  with  the  location  in  s  of
least  two  children;;  i.e.,  it  has  degree  at  least  3. the  suffix  ending  at  that  leaf.
The  substring  labels  for  the  edges  leading  from  a  node  to
its  children  must  begin  with  different  symbols.
By  concatenating  the  substrings  along  edges,  each  path  from  the  root  to  a  leaf  corresponds  to  a
unique  suffix  of   .

See  Figure  1  for  an  example  of  a  suffix  tree.

Given:  A  DNA  string    (of  length  at  most  20  kbp)  with   $  appended,  a  positive  integer   ,  and  a  list
of  edges  defining  the  suffix  tree  of   .  Each  edge  is  represented  by  four  components:

1.   the  label  of  its  parent  node  in   ;;

 2.   the  label  of  its  child  node  in   ;;
3.   the  location  of  the  substring    of    assigned  to  the  edge;;  and
4.   the  length  of   .

Return:  The  longest  substring  of    that  occurs  at  least    times  in   .  (If  multiple  solutions  exist,  you
may  return  any  single  solution.)

Sample  Dataset

CATACATAC$
2
node1  node2  1  1
node1  node7  2  1
node1  node14  3  3
node1  node17  10  1
node2  node3  2  4
node2  node6  10  1
node3  node4  6  5
node3  node5  10  1
node7  node8  3  3
node7  node11  5  1
node8  node9  6  5
node8  node10  10  1
node11  node12  6  5
node11  node13  10  1
node14  node15  6  5
node14  node16  10  1

Sample  Output

CATAC

Hint

How  can  repeated  substrings  of    be  located  in   ?

Problem  68
Newick  Format  with  Edge  Weights

Weighting  the  Tree

A  vital  goal  of  creating  phylogenies  is  to  quantify  a  molecular  clock  that  indicates  the  amount  of
evolutionary  time  separating  two  members  of  the  phylogeny.  To  this  end,  we  will  assign  numbers  to
the  edges  of  a  tree  so  that  the  number  assigned  to  an  edge  represents  the  amount  of  time
 separating  the  two  species  at  each  end  of  the  edge.  More  generally,  the
evolutionary  time  between  any  two  species  will  be  given  by  simply  adding  the
individual  times  connecting  the  nodes.

Problem

In  a  weighted  tree,  each  edge  is  assigned  a  (usually  positive)  number,  called  its  weight.  The
distance  between  two  nodes  in  a  weighted  tree  becomes  the  sum  of  the  weights  along  the  unique  path
connecting  the  nodes.

To  generalize  Newick  format  to  the  case  of  a  weighted  tree   ,  during  our  repeated  "key  step,"  if  leaves
 are  neighbors  in   ,  and  all  these  leaves  are  incident  to   ,  then  we  replace    with  
,  where    is  now  the  weight  on  the  edge   .

Given:  A  collection  of    weighted  trees  ( )  in  Newick  format,  with  each  tree  containing  at
most  200  nodes;;  each  tree    is  followed  by  a  pair  of  nodes    and    in   .

Return:  A  collection  of    numbers,  for  which  the   th  number  represents  the  distance  between  
and    in   .

Sample  Dataset

(dog:42,cat:33);
cat  dog

((dog:4,cat:3):74,robot:98,elephant:58);
dog  elephant

Sample  Output

75  136

Problem  69
Wobble  Bonding  and  RNA  Secondary  Structures

Don't  Look  Down

We  have  discussed  the  problem  of  counting  RNA  secondary  structures  in
previous  problems.  In  this  problem,  we  will  add  some  assumptions  to  those  used
in  “Motzkin  Numbers  and  RNA  Secondary  Structures”  to  provide  ourselves  with  an
ultimately  robust  way  of  counting  feasible  RNA  secondary  structures.

First,  in  addition  to  the  classic  Watson  and  Crick  base  pairing  of  adenine  with  uracil  and  cytosine
with  guanine,  uracil  sometimes  bonds  with  guanine,  forming  what  is  called  a  wobble  base  pair.
As  a  result,  we  would  like  to  allow  wobble  base  pairing.
 Second,  although  RNA  folds  over  itself  during  base  pairing,  the  structure  of  the  molecule  is  rigid
enough  to  prevent  bases  located  very  close  to  each  other  on  the  strand  from  bonding  to  each  other.

Problem

Given  an  RNA  string   ,  we  will  augment  the  bonding  graph  of  
 by  adding  basepair  edges  connecting  all  occurrences  of  'U'
to  all  occurrences  of  'G'  in  order  to  represent  possible  wobble
base  pairs.

We  say  that  a  matching  in  the  bonding  graph  for    is  valid  if  it
is  noncrossing  (to  prevent  pseudoknots)  and  has  the  property
that  a  basepair  edge  in  the  matching  cannot  connect  symbols  
Figure  1.  A  valid  matching  of
 and    unless    (to  prevent  nearby  nucleotides basepair  edges  in  the  bonding
from  base  pairing). graph  of  an  RNA  string,  followed  by
a  diagram  of  how  this  bonding  will
See  Figure  1  for  an  example  of  a  valid  matching  if  we  allow induce  the  resulting  folded  RNA.
wobble  base  pairs.  In  this  problem,  we  will  wish  to  count  all
possible  valid  matchings  in  a  given  bonding  graph;;  see  Figure
2  for  all  possible  valid  matchings  in  a  small  bonding  graph,
assuming  that  we  allow  wobble  base  pairing.

Given:  An  RNA  string    (of  length  at  most  200  bp).
Return:  The  total  number  of  distinct  valid  matchings  of
basepair  edges  in  the  bonding  graph  of   .  Assume  that
wobble  base  pairing  is  allowed.
Figure  2.  All  12  possible  valid
basepair  matchings  in  the  bonding
Sample  Dataset graph  corresponding  to  the  string  s
=  CGAUGCUAG  (including  the  trivial
matching  in  which  no  edges  are
AUGCUAGUACGGAGCGAGUCUAGCGAGCGAUGUCGUGAGUAC matched.)  Courtesy  Brian  Tjaden.
UAUAUAUGCGCAUAAGCCACGU

Sample  Output

284850219977421

Problem  70
Counting  Disease  Carriers

Genetic  Drift  and  the  Hardy-­Weinberg  Principle

Mendel's  laws  of  segration  and  independent  assortment  are  excellent  for  the
study  of  individual  organisms  and  their  progeny,  but  they  say  nothing  about  how
alleles  move  through  a  population  over  time.  Our  first  question  is:  when  can  we
assume  that  the  ratio  of  an  allele  in  a  population,  called  the  allele  frequency,  is
 stable?

G.  H.  Hardy  and  Wilhelm  Weinberg  independently  considered  this  question  at  the  turn  of  the  20th
Century,  shortly  after  Mendel's  ideas  had  been  rediscovered.  They  concluded  that  the  percentage  of
an  allele  in  a  population  of  individuals  is  in  genetic  equilibrium  when  five  conditions  are  satisfied:

1.   The  population  is  so  large  that  random  changes  in  the  allele  frequency  are  negligible.
2.   No  new  mutations  are  affecting  the  gene  of  interest;;
3.   The  gene  does  not  influence  survival  or  reproduction,  so  that  natural  selection  is  not  occurring;;
4.   Gene  flow,  or  the  change  in  allele  frequency  due  to  migration  into  and  out  of  the  population,  is
negligible.
5.   Mating  occurs  randomly  with  respect  to  the  gene  of  interest.

The  Hardy-­Weinberg  principle  states  that  if  a  population  is  in  genetic  equilibrium  for  a  given
allele,  then  its  frequency  will  remain  constant  and  evenly  distributed  through  the  population.  Unless
the  gene  in  question  is  important  to  survival  or  reproduction,  Hardy-­Weinberg  usually  offers  a
reasonable  enough  model  of  population  genetics.

One  of  the  many  benefits  of  the  Mendelian  theory  of  inheritance  and  simplifying  models  like  Hardy-­
Weinberg  is  that  they  help  us  predict  the  probability  with  which  genetic  diseases  will  be  inherited,
so  as  to  take  appropriate  preventative  measures.  Genetic  diseases  are  usually  caused  by
mutations  to  chromosomes,  which  are  passed  on  to  subsequent  generations.

The  simplest  and  most  widespread  case  of  a  genetic  disease  is  a  single  gene  disorder,  which  is
caused  by  a  single  mutated  gene.  Over  4,000  such  human  diseases  have  been  identified,  including
cystic  fibrosis  and  sickle-­cell  anemia.  In  both  of  these  cases,  the  individual  must  possess  two
recessive  alleles  for  a  gene  in  order  to  contract  the  disease.  Thus,  carriers  can  live  their  entire  lives
without  knowing  that  they  can  pass  the  disease  on  to  their  children.

The  above  introduction  to  genetic  equilibrium  leaves  us  with  a  basic  and  yet  very  practical  question
regarding  gene  disorders:  if  we  know  the  number  of  people  who  have  a  disease  encoded  by  a
recessive  allele,  can  we  predict  the  number  of  carriers  in  the  population?

Problem

To  model  the  Hardy-­Weinberg  principle,  assume  that  we  have  a  population  of    diploid  individuals.  If
an  allele  is  in  genetic  equilibrium,  then  because  mating  is  random,  we  may  view  the  
chromosomes  as  receiving  their  alleles  uniformly.  In  other  words,  if  there  are    dominant  alleles,  then
the  probability  of  a  selected  chromosome  exhibiting  the  dominant  allele  is  simply   .

Because  the  first  assumption  of  genetic  equilibrium  states  that  the  population  is  so  large  as  to  be
ignored,  we  will  assume  that    is  infinite,  so  that  we  only  need  to  concern  ourselves  with  the  value  of  
.

Given:  An  array    for  which    represents  the  proportion  of  homozygous  recessive  individuals  for
the   -­th  Mendelian  factor  in  a  diploid  population.  Assume  that  the  population  is  in  genetic  equilibrium
for  all  factors.

Return:  An  array    having  the  same  length  as    in  which    represents  the  probability  that  a
randomly  selected  individual  carries  at  least  one  copy  of  the  recessive  allele  for  the   -­th  factor.

Sample  Dataset

0.1  0.25  0.5

 Sample  Output

0.532  0.75  0.914

Problem  71
Creating  a  Character  Table  from  Genetic  Strings

Phylogeny  from  Genetic  Characters

In  “Creating  a  Character  Table”,  we  introduced  the  character  table  as  a  way  of
representing  a  number  of  characters  simultaneously.  In  that  problem,  we  found  a
character  table  representing  an  unrooted  binary  tree  on  a  collection  of  taxa.

Of  course,  in  practice,  the  problem  operates  in  reverse.  We  first  need  to  generate  a  character  table
before  we  can  model  a  phylogeny  on  this  table.  In  modern  genetics,  a  reliable  way  to  obtain  a  large
number  of  characters  is  by  using  SNPs.  As  mentioned  in  “Counting  Subsets”,  for  a  given  SNP,  we
can  divide  taxa  into  two  sets  depending  on  which  of  two  bases  is  present  at  the  nucleotide,  thus
defining  the  split  of  a  character.

Problem

A  collection  of  strings  is  characterizable  if  there  are  at  most  two  possible  choices  for  the  symbol  at
each  position  of  the  strings.

Given:  A  collection  of  at  most  100  characterizable  DNA  strings,  each  of  length  at  most  300  bp.
Return:  A  character  table  for  which  each  nontrivial  character  encodes  the  symbol  choice  at  a  single
position  of  the  strings.  (Note:  the  choice  of  assigning  '1'  and  '0'  to  the  two  states  of  each  SNP  in  the
strings  is  arbitrary.)

Sample  Dataset

ATGCTACC
CGTTTACC
ATTCGACC
AGTCTCCC
CGTCTATC

Sample  Output

10110
10100
 Note

Recall  that  the  character  table  does  not  encode  trivial  characters.

Problem  72
Counting  Optimal  Alignments

Beware  of  Alignment  Inference

In  “Edit  Distance  Alignment”,  we  introduced  the  concept  of  an  alignment  of  two
genetic  strings  having  differing  lengths  with  respect  to  edit  distance.  This  provided
us  with  a  way  of  visualizing  a  specific  collection  of  symbol  substitutions,
insertions,  and  deletions  that  could  have  taken  place  on  the  evolutionary  path
between  the  two  strings.

However,  simply  finding  one  optimal  alignment  and  declaring  that  it  represents  a  true  evolutionary
history  is  a  dangerous  idea  because  the  actual  evolutionary  picture  may  be  suboptimal.  For  that
matter,  the  collection  of  all  optimal  alignments  may  be  huge,  and  the  characteristics  of  these
alignments  could  differ  widely.

In  order  to  begin  analyzing  the  collection  of  optimal  alignments  for  a  pair  of  strings,  the  first
question  we  will  ask  is  simple:  just  how  many  optimal  alignments  exist?

Problem

Recall  from  “Edit  Distance  Alignment”  that  if    and    are  the  augmented  strings  corresponding  to  an
alignment  of  strings    and   ,  then  the  edit  alignment  score  of    and    was  given  by  the  Hamming
distance    (because    and    have  the  same  length  and  already  include  gap  symbols  to
denote  insertions/deletions).

As  a  result,  we  obtain   ,  where  the  minimum  is  taken  over  all
alignments  of    and   .  Strings    and    achieving  this  minimum  correspond  to  an  optimal  alignment
with  respect  to  edit  alignment  score.

Given:  Two  protein  strings    and    in  FASTA  format,  each  of  length  at  most  1000  aa.
Return:  The  total  number  of  optimal  alignments  of    and    with  respect  to  edit  alignment  score,
modulo  134,217,727  (227-­1).

Sample  Dataset

>Rosalind_78
PLEASANTLY
>Rosalind_33
MEANLY
 Sample  Output

Why  Are  We  Counting  Modulo  134,217,727?

As  a  simple  example,  say  that  we  attempt  to  count  some  statistic  modulo  10.  If  computing  the
statistic  requires  us  to  multiply  a  collection  of  integers,  and  at  any  point  we  multiply  by  a  multiple
of  10,  then  the  statistic  will  automatically  become  a  multiple  of  10  and  thus  congruent  to  0  modulo
10.  A  similar  issue  can  arise  when  we  count  a  huge  number  modulo  any  composite  number;;
however,  if  we  count  modulo  a  large  prime  number    (i.e.,  one  without  any  divisors  other  than
itself),  then  problems  can  only  ever  arise  if  when  counting  our  statistic,  we  multiply  by  a  multiple  of  
.

Problem  73
Counting  Unrooted  Binary  Trees

Counting  Trees

A  natural  question  is  to  be  able  to  count  the  total  number  of  distinct  unrooted
binary  trees  having    leaves,  where  each  leaf  is  labeled  by  some  taxon.  Before
we  can  count  all  these  trees,  however,  we  need  to  have  a  notion  of  when  two  such
trees  are  the  same.

Our  tool  will  be  the  split.  Recall  from  “Creating  a  Character  Table”  that  removing  any  edge  from  a
tree    separates  its  leaves  into  sets    and   ,  so  that  each  edge  of    can  be  labeled  by  this
split   .  As  a  result,  an  unrooted  binary  tree  can  be  represented  uniquely  by  its  collection  of
splits.

Problem

Two  unrooted  binary  trees    and    having  the  same    labeled  leaves  are  considered  to  be
equivalent  if  there  is  some  assignment  of  labels  to  the  internal  nodes  of    and    so  that  the
adjacency  lists  of  the  two  trees  coincide.  As  a  result,  note  that    and    must  have  the  same  splits;;
conversely,  if  the  two  trees  do  not  have  the  same  splits,  then  they  are  considered  distinct.

Let    denote  the  total  number  of  distinct  unrooted  binary  trees  having    labeled  leaves.

Given:  A  positive  integer    ( ).

Return:  The  value  of    modulo  1,000,000.

Sample  Dataset

5
 Sample  Output

15

Problem  74
Global  Alignment  with  Scoring  Matrix

Generalizing  the  Alignment  Score

The  edit  alignment  score  in  “Edit  Distance  Alignment”  counted  the  total  number  of
edit  operations  implied  by  an  alignment;;  we  could  equivalently  think  of  this  scoring
function  as  assigning  a  cost  of  1  to  each  such  operation.  Another  common
scoring  function  awards  matched  symbols  with  1  and  penalizes
substituted/inserted/deleted  symbols  equally  by  assigning  each  one  a  score  of  0,  so  that  the
maximum  score  of  an  alignment  becomes  the  length  of  a  longest  common  subsequence  of    and  
(see  “Finding  a  Shared  Spliced  Motif”).  In  general,  the  alignment  score  is  simply  a  scoring
function  that  assigns  costs  to  edit  operations  encoded  by  the  alignment.

One  natural  way  of  adding  complexity  to  alignment  scoring  functions  is  by  changing  the  alignment
score  based  on  which  symbols  are  substituted;;  many  methods  have  been  proposed  for  doing  this.
Another  way  to  do  so  is  to  vary  the  penalty  assigned  to  the  insertion  or  deletion  of  symbols.

In  general,  alignment  scores  can  be  either  maximized  or  minimized  depending  on  how  scores  are
established.  The  general  problem  of  optimizing  a  particular  alignment  score  is  called  global
alignment.

Problem

To  penalize  symbol  substitutions  differently  depending  on  which  two  symbols  are  involved  in  the
substitution,  we  obtain  a  scoring  matrix    in  which    represents  the  (negative)  score  assigned  to
a  substitution  of  the   th  symbol  of  our  alphabet    with  the   th  symbol  of   .

A  gap  penalty  is  the  component  deducted  from  alignment  score  due  to  the  presence  of  a  gap.  A  gap
penalty  may  be  a  function  of  the  length  of  the  gap;;  for  example,  a  linear  gap  penalty  is  a  constant  
such  that  each  inserted  or  deleted  symbol  is  charged   ;;  as  a  result,  the  cost  of  a  gap  of  length    is
equal  to   .

Given:  Two  protein  strings    and    in  FASTA  format  (each  of  length  at  most  1000  aa).
Return:  The  maximum  alignment  score  between    and   .  Use:
The  BLOSUM62  scoring  matrix.
Linear  gap  penalty  equal  to  5  (i.e.,  a  cost  of  -­5  is  assessed  for  each  gap  symbol).

Sample  Dataset
 >Rosalind_67
PLEASANTLY
>Rosalind_17
MEANLY

Sample  Output

Problem  75
Genome  Assembly  with  Perfect  Coverage

Cyclic  Chromosomes

Recall  that  although  chromosomes  taken  from  eukaryotes  have  a  linear  structure,
many  bacterial  chromosomes  are  actually  circular.  We  represented  a  linear
chromosome  with  a  DNA  string,  so  we  only  need  to  modify  the  definition  of  string
to  model  circular  chromosomes.

Perfect  coverage  is  the  phenomenon  in  fragment  assembly  of  having  a  read  (or   -­mer)  begin  at
every  possible  location  in  the  genome.  Unfortunately,  perfect  coverage  is  still  difficult  to  achieve,  but
fragment  assembly  technology  continues  to  improve  by  leaps  and  bounds,  and  perfect  coverage  is
perhaps  not  the  fantasy  it  once  was.

Problem

A  circular  string  is  a  string  that  does  not  have  an  initial  or  terminal  element;;  instead,  the  string  is
viewed  as  a  necklace  of  symbols.  We  can  represent  a  circular  string  as  a  string  enclosed  in
parentheses.  For  example,  consider  the  circular  DNA  string  (ACGTAC),  and  note  that  because  the
string  "wraps  around"  at  the  end,  this  circular  string  can  equally  be  represented  by  (CGTACA),
(GTACAC),  (TACACG),  (ACACGT),  and  (CACGTA).  The  definitions  of  substrings  and  superstrings  are
easy  to  generalize  to  the  case  of  circular  strings  (keeping  in  mind  that  substrings  are  allowed  to  wrap
around).

Given:  A  collection  of  (error-­free)  DNA   -­mers  ( )  taken  from  the  same  strand  of  a  circular
chromosome.  In  this  dataset,  all   -­mers  from  this  strand  of  the  chromosome  are  present,  and  their  de
Bruijn  graph  consists  of  exactly  one  simple  cycle.

Return:  A  cyclic  superstring  of  minimal  length  containing  the  reads  (thus  corresponding  to  a
candidate  cyclic  chromosome).

Sample  Dataset

ATTAC
TACAG
 GATTA
ACAGA
CAGAT
TTACA
AGATT

Sample  Output

GATTACA

Note

The  assumption  made  above  that  all  reads  derive  from  the  same  strand  is  practically  unrealistic;;  in
reality,  researchers  will  not  know  the  strand  of  DNA  from  which  a  given  read  has  been  sequenced.

Problem  76
Matching  a  Spectrum  to  a  Protein

Searching  the  Protein  Database

Many  proteins  have  already  been  identified  for  a  wide  variety  of  organisms.
Accordingly,  there  are  a  large  number  of  protein  databases  available,  and  so  the
first  step  after  creating  a  mass  spectrum  for  an  unidentified  protein  is  to  search
through  these  databases  for  a  known  protein  with  a  highly  similar  spectrum.  In
this  manner,  many  similar  proteins  found  in  different  species  have  been  identified,  which  aids
researchers  in  determining  protein  function.

In  “Comparing  Spectra  with  the  Spectral  Convolution”,  we  introduced  the  spectral  convolution  and
used  it  to  measure  the  similarity  of  simplified  spectra.  In  this  problem,  we  would  like  to  extend  this
idea  to  find  the  most  similar  protein  in  a  database  to  a  spectrum  taken  from  an  unknown  protein.
Our  plan  is  to  use  the  spectral  convolution  to  find  the  largest  possible  number  of  masses  that  each
database  protein  shares  with  our  candidate  protein  after  shifting,  and  then  select  the  database
protein  having  the  largest  such  number  of  shared  masses.

Problem

The  complete  spectrum  of  a  weighted  string    is  the  multiset    containing  the  weights  of  every
prefix  and  suffix  of   .

Given:  A  positive  integer    followed  by  a  collection  of    protein  strings  

,   ,  ...,    and  a
multiset    of  positive  numbers  (corresponding  to  the  complete  spectrum  of  some  unknown  protein
string).

Return:  The  maximum  multiplicity  of    taken  over  all  strings   ,  followed  by  the  string  
 for  which  this  maximum  multiplicity  occurs  (you  may  output  any  such  value  if  multiple  solutions
 exist).

Sample  Dataset

4
GSDMQS
VWICN
IASWMQS
PVSMGAD
445.17838
115.02694
186.07931
314.13789
317.1198
215.09061

Sample  Output

3
IASWMQS

Problem  77
Quartets

Incomplete  Characters

The  modern  revolution  in  genome  sequencing  has  produced  a  huge  amount  of
genetic  data  for  a  wide  variety  of  species.  One  ultimate  goal  of  possessing  all  this
information  is  to  be  able  to  construct  complete  phylogenies  via  direct  genome
analysis.

For  example,  say  that  we  have  a  gene  shared  by  a  number  of  taxa.  We  could  create  a  character
based  on  whether  species  are  known  to  possess  the  gene  or  not,  and  then  use  a  huge  character
table  to  construct  our  desired  phylogeny.  However,  the  present  bottleneck  with  such  a  method  is
that  it  assumes  that  we  already  possess  complete  genome  information  for  all  possible  species.
The  race  is  on  to  sequence  as  many  species  genomes  as  possible;;  for  instance,  the  Genome  10K
Project  aims  to  sequence  10,000  species  genomes  over  the  next  decade.  Yet  for  the  time  being,
possessing  a  complete  genomic  picture  of  all  Earth's  species  remains  a  dream.

As  a  result  of  these  practical  limitations,  we  need  to  be  able  to  work  with  partial  characters,
which  divide  taxa  into  three  separate  groups:  those  possessing  the  character,  those  not
possessing  the  character,  and  those  for  which  we  do  not  yet  have  conclusive  information.

Problem
 A  partial  split  of  a  set    of    taxa  models  a  partial  character  and  is  denoted  by   ,  where    and
 are  still  the  two  disjoint  subsets  of  taxa  divided  by  the  character.  Unlike  in  the  case  of  splits,  we  do
not  necessarily  require  that   ;;    corresponds  to  those  taxa  for  which  we  lack
conclusive  evidence  regarding  the  character.

We  can  assemble  a  collection  of  partial  characters  into  a  generalized  partial  character  table    in
which  the  symbol    is  placed  in    if  we  do  not  have  conclusive  evidence  regarding  the   th  taxon
with  respect  to  the   th  partial  character.

A  quartet  is  a  partial  split    in  which  both    and    contain  precisely  two  elements.  For  the
sake  of  simplicity,  we  often  will  consider  quartets  instead  of  partial  characters.  We  say  that  a  quartet  
 is  inferred  from  a  partial  split    if    and    (or  equivalently    and  
).  For  example,    and    can  be  inferred  from   .

Given:  A  partial  character  table   .

Return:  The  collection  of  all  quartets  that  can  be  inferred  from  the  splits  corresponding  to  the
underlying  characters  of   .

Sample  Dataset

cat  dog  elephant  ostrich  mouse  rabbit  robot

01xxx00
x11xx00
111x00x

Sample  Output

{elephant,  dog}  {rabbit,  robot}

{cat,  dog}  {mouse,  rabbit}
{mouse,  rabbit}  {cat,  elephant}
{dog,  elephant}  {mouse,  rabbit}

Problem  78
Using  the  Spectrum  Graph  to  Infer  Peptides

Getting  Real  with  Spectra

In  “Inferring  Peptide  from  Full  Spectrum”,  we  considered  an  idealized  version  of  the
simplified  spectrum  in  which  every  cut  through  a  given  peptide  was  produced,  so
that  the  spectrum  possessed  all  possible  b-­ions  and  y-­ions  cutting  the  peptide.  In
reality,  not  every  cut  will  be  produced  in  a  spectrum,  which  may  also  contain
errors.  As  a  result,  it  is  difficult  or  impossible  to  recover  an  entire  peptide  from  a  single  spectrum.

In  the  more  practical  case  of  a  mass  spectrum,  where  intensity  is  plotted  against  ions'  mass-­
charge  ratios,  inferring  the  protein  is  also  greatly  complicated  by  the  presence  of  erratic  peaks  in
the  spectrum.
 Problem

For  a  weighted  alphabet    and  a  collection    of  positive  real  numbers,  the  spectrum  graph  of    is
a  digraph  constructed  in  the  following  way.  First,  create  a  node  for  every  real  number  in   .  Then,
connect  a  pair  of  nodes  with  a  directed  edge    if    and    is  equal  to  the  weight  of  a
single  symbol  in   .  We  may  then  label  the  edge  with  this  symbol.

In  this  problem,  we  say  that  a  weighted  string    matches    if  there  is  some
increasing  sequence  of  positive  real  numbers    in    such  that  
,   ,  ...,  and   .

Given:  A  list    (of  length  at  most  100)  containing  positive  real  numbers.
Return:  The  longest  protein  string  that  matches  the  spectrum  graph  of    (if  multiple  solutions
exist,  you  may  output  any  one  of  them).  Consult  the  monoisotopic  mass  table.

Sample  Dataset

3524.8542
3623.5245
3710.9335
3841.974
3929.00603
3970.0326
4026.05879
4057.0646
4083.08025

Sample  Output

WMSPG

Hint

How  can  our  question  be  rephrased  in  terms  of  the  spectrum  graph?

Problem  79
Encoding  Suffix  Trees

Creating  a  Suffix  Tree

In  “Finding  the  Longest  Multiple  Repeat”,  we  introduced  the  suffix  tree.  This  data
structure  has  a  wide  array  of  applications,  one  of  which  was  to  help  us  identify
long  repeats  in  a  genome.  In  that  problem,  we  provided  the  tree  as  part  of  the
 dataset,  but  a  vital  computational  exercise  is  to  create  the  suffix  tree  solely  from  a  string.

Problem

Given  a  string    having  length   ,  recall  that  its  suffix  tree  

 is  defined  by  the  following  properties:

 is  a  rooted  tree  having  exactly    leaves.

Every  edge  of    is  labeled  with  a  substring  of   ,
where    is  the  string  formed  by  adding  a  placeholder
symbol   $  to  the  end  of   . Figure  1.  The  suffix  tree  for  s  =
Every  internal  node  of    other  than  the  root  has  at GTCCGAAGCTCCGG.  Note  that  the
dollar  sign  has  been  appended  to  a
least  two  children;;  i.e.,  it  has  degree  at  least  3. substring  of  the  tree  to  mark  the  end
The  substring  labels  for  the  edges  leading  down  from  a of  s.  Every  path  from  the  root  to  a  leaf
corresponds  to  a  unique  suffix  of
node  to  its  children  must  begin  with  different  symbols. GTCCGAAGCTCCGG,  and  each  leaf
By  concatenating  the  substrings  along  edges,  each  path is  labeled  with  the  location  in  s  of
from  the  root  to  a  leaf  corresponds  to  a  unique  suffix  of   the  suffix  ending  at  that  leaf.
.

Figure  1  contains  an  example  of  a  suffix  tree.

Given:  A  DNA  string    of  length  at  most  1kbp.

Return:  The  substrings  of    encoding  the  edges  of  the  suffix  tree  for   .  You  may  list  these
substrings  in  any  order.

Sample  Dataset

ATAAATG$

Sample  Output

AAATG$
G$
T
ATG$
TG$
A
A
AAATG$
G$
T
G$
$

Problem  80
Character-­Based  Phylogeny
 Introduction  to  Character-­Based  Phylogeny

In  “Creating  a  Character  Table”,  we  discussed  the  construction  of  a  character

table  from  a  collection  of  characters  represented  by  subsets  of  our  taxa.  However,
the  ultimate  goal  is  to  be  able  to  construct  a  phylogeny  from  this  character  table.

The  issues  at  hand  are  that  we  want  to  ensure  that  we  have  enough  characters  to  actually
construct  a  phylogeny,  and  that  our  characters  do  not  conflict  with  each  other.

Problem

Because  a  tree  having    nodes  has    edges  (see  “Completing  a  Tree”),  removing  a  single  edge
from  a  tree  will  produce  two  smaller,  disjoint  trees.  Recall  from  “Creating  a  Character  Table”  that  for
this  reason,  each  edge  of  an  unrooted  binary  tree  corresponds  to  a  split   ,  where    is  a  subset
of  the  taxa.

A  consistent  character  table  is  one  whose  characters'  splits  do  not  conflict  with  the  edge  splits  of
some  unrooted  binary  tree    on  the    taxa.  More  precisely,    conflicts  with    if  all
four  intersections   ,   ,   ,  and    are  nonempty.  As  a  simple
example,  consider  the  conflicting  splits    and   .

More  generally,  given  a  consistent  character  table   ,  an  unrooted  binary  tree    "models"    if  the
edge  splits  of    agree  with  the  splits  induced  from  the  characters  of   .

Given:  A  list  of    species  ( )  and  an   -­column  character  table    in  which  the   th  column
denotes  the   th  species.

Return:  An  unrooted  binary  tree  in  Newick  format  that  models   .

Sample  Dataset

cat  dog  elephant  mouse  rabbit  rat

011101
001101
001100

Sample  Output

(dog,(cat,rabbit),(rat,(elephant,mouse)));

Problem  81
Counting  Quartets

Introduction  to  Quartet-­Based  Phylogeny

 In  “Quartets”,  we  introduced  partial  splits  modeling  partial  characters  on  a
collection  of  taxa.  Our  aim  is  to  use  the  quartets  inferred  from  partial  splits  to
construct  a  phylogeny  on  the  taxa.  This  procedure  is  called  quartet-­based
phylogeny.

We  could  construct  a  phylogeny  directly  from  a  collection  of  partial  splits,  but  it  is
not  immediately  clear  how  many  different  splits  we  would  need.  Hence,  our  first
question  is  to  ask  how  many  quartets  are  required  to  be  able  to  infer  a  tree;;  in  this
problem  we  will  ask  the  reverse  question  of  how  many  quartets  can  be  inferred  from  a  known  tree.

Problem

A  quartet    is  consistent  with  a  binary  tree    if  the  quartet  can  be  inferred  from  one  of  the
splits  of    (see  “Quartets”  for  a  description  of  inferring  quartets  from  splits).

Let    denote  the  total  number  of  quartets  that  are  consistent  with   .

Given:  A  positive  integer    ( ),  followed  by  an  unrooted  binary  tree    in  Newick
format  on    taxa.

Return:  The  value  of    modulo  1,000,000.

Sample  Dataset

6
(lobster,(cat,dog),(caterpillar,(elephant,mouse)));

Sample  Output

15

Problem  82
Enumerating  Unrooted  Binary  Trees

Seeing  the  Forest

In  “Counting  Unrooted  Binary  Trees”,  we  found  a  way  to  count  the  number  of
unrooted  binary  trees  representing  phylogenies  on    taxa.  Our  observation  was
that  two  such  trees  are  considered  distinct  when  they  do  not  share  the  same
collection  of  splits.

Counting  all  these  trees  is  one  task,  but  actually  understanding  how  to  write  them  out  in  a  list  (i.e.,
enumerating  them)  is  another,  which  will  be  the  focus  of  this  problem.

Problem
 Recall  the  definition  of  Newick  format  from  “Distances  in
Trees”  as  a  way  of  encoding  trees.  See  Figure  1  for  an
example  of  Newick  format  applied  to  an  unrooted  binary  tree
whose  five  leaves  are  labeled  (note  that  the  same  tree  can
have  multiple  Newick  representations).

Given:  A  collection  of  species  names  representing    taxa.

Return:  A  list  containing  all  unrooted  binary  trees  whose Figure  1.  This  unrooted  binary  tree
may  be  represented  in  Newick
leaves  are  these    taxa.  Trees  should  be  given  in  Newick format  by  (((a,b),c),(d,e));;  another
format,  with  one  tree  on  each  line;;  the  order  of  the  trees  is way  of  encoding  it  is  ((a,b),(c,(d,e))).
unimportant.

Sample  Dataset

dog  cat  mouse  elephant

Sample  Output

(((mouse,cat),elephant))dog;
(((elephant,mouse),cat))dog;
(((elephant,cat),mouse))dog;

Problem  83
Genome  Assembly  Using  Reads

Putting  the  Puzzle  Together

In  practical  genome  sequencing,  even  if  we  assume  that  reads  have  been
sequenced  without  errors,  we  have  no  idea  of  knowing  immediately  the  particular
strand  of  DNA  a  read  has  come  from.

Also,  our  reads  may  not  have  the  same  length.  In  1995,  Idury  and  Waterman  proposed  a  way  to
boost  read  coverage  and  achieve  uniform  read  length  by  breaking  long  reads  into  overlapping   -­
mers  for  some  fixed  value  of   .  For  example,  a  100  bp  read  could  be  split  into  51  overlapping  50-­
mers.

Problem

A  directed  cycle  is  simply  a  cycle  in  a  directed  graph  in  which  the  head  of  one  edge  is  equal  to  the
tail  of  the  next  (so  that  every  edge  in  the  cycle  is  traversed  in  the  same  direction).

For  a  set  of  DNA  strings    and  a  positive  integer   ,  let    denote  the  collection  of  all  possible   -­mers
of  the  strings  in   .

Given:  A  collection    of  (error-­free)  reads  of  equal  length  (not  exceeding  50  bp).  In  this  dataset,  for
 some  positive  integer   ,  the  de  Bruijn  graph    on    consists  of  exactly  two  directed
cycles.

Return:  A  cyclic  superstring  of  minimal  length  containing  every  read  or  its  reverse  complement.

Sample  Dataset

AATCT
TGTAA
GATTA
ACAGA

Sample  Output

GATTACA

Note

The  reads  "AATCT"  and  "TGTAA"  are  not  present  in  the  answer,  but  their  reverse  complements
"AGATT"  and  "TTACA"  are  present  in  the  circular  string  (GATTACA).

Problem  84
Global  Alignment  with  Constant  Gap  Penalty

Penalizing  Large  Insertions  and  Deletions

In  dealing  with  global  alignment  in  “Global  Alignment  with  Scoring  Matrix”,  we
encountered  a  linear  gap  penalty,  in  which  the  insertion  or  deletion  of  a  gap  is
penalized  by  some  constant  times  the  length  of  the  gap.  However,  this  model  is
not  necessarily  the  most  practical  model,  as  one  large  rearrangement  could  have
inserted  or  deleted  a  long  gap  in  a  single  step  to  transform  one  genetic  string  into  another.

Problem

In  a  constant  gap  penalty,  every  gap  receives  some  predetermined  constant  penalty,  regardless  of  its
length.  Thus,  the  insertion  or  deletion  of  1000  contiguous  symbols  is  penalized  equally  to  that  of  a
single  symbol.

Given:  Two  protein  strings    and    in  FASTA  format  (each  of  length  at  most  1000  aa).
Return:  The  maximum  alignment  score  between    and   .  Use:
The  BLOSUM62  scoring  matrix.
Constant  gap  penalty  equal  to  5.
 Sample  Dataset

>Rosalind_79
PLEASANTLY
>Rosalind_41
MEANLY

Sample  Output

13

Problem  85
Linguistic  Complexity  of  a  Genome

Getting  Repetitive

We  have  seen  that  every  genome  contains  a  large  number  of  repeats  and  noted
that  the  Alu  repeat  recurs  around  a  million  times  on  the  human  genome.  Yet
exactly  how  repetitive  is  the  human  genome?

To  frame  such  a  vague  question  mathematically,  we  first  need  to  make  the  observation  that  if  the
genome  were  formed  by  adding  nucleobases  randomly,  with  each  base  having  a  1/4  probability  of
being  added  at  each  nucleotide  position,  then  we  should  expect  to  see  a  huge  number  of  different
substrings  in  the  genome.  Yet  (to  take  a  simple  case)  the  genome  containing  only  adenosine  and
forming  the  DNA  string  "AAAAAA...AAA"  has  relatively  very  few  distinct  substrings.

Now,  real  genomes  are  formed  by  a  process  that  chooses  nucleotides  somewhere  in  between
these  two  extreme  cases,  and  so  to  quantify  just  how  random  this  process  is,  we  need  to  take  the
percentage  of  distinct  substrings  appearing  in  a  genome  with  respect  to  the  maximum  possible
number  of  distinct  substrings  that  could  appear  in  a  genome  of  the  same  length.

Problem

Given  a  length    string    formed  over  an  alphabet    of  size   ,  let  the  "substring  count"  
denote  the  total  number  of  distinct  substrings  of   .  Furthermore,  let  the  "maximum  substring  count"  
 denote  the  maximum  number  of  distinct  substrings  that  could  appear  in  a  string  of  length  
formed  over   .

The  linguistic  complexity  of    (written   )  is  equal  to   ;;  in  other  words,    represents
the  percentage  of  observed  substrings  of    to  the  total  number  that  are  theoretically  possible.  Note
that   ,  with  smaller  values  of    indicating  that    is  more  repetitive.

As  an  example,  consider  the  DNA  string  ( )   .  In  the  following  table,  we
demonstrate  that    by  considering  the  number  of  observed  and  possible  length  
 substrings  of   ,  which  are  denoted  by    and   ,  respectively.  (Observe  that  
 and   .)

1 3 4
2 5 8
3 6 7
4 6 6
5 5 5
6 4 4
7 3 3
8 2 2
9 1 1
Total 35 40

Given:  A  DNA  string    of  length  at  most  100  kbp.

Return:  The  linguistic  complexity   .

Sample  Dataset

ATTTGGATT

Sample  Output

0.875

Hint

Why  does  this  problem  follow  “Encoding  Suffix  Trees”?

Problem  86
Local  Alignment  with  Scoring  Matrix

Aligning  Similar  Substrings

Whereas  global  alignment  (see  “Global  Alignment  with  Scoring  Matrix”)  can  be
helpful  for  comparing  genetic  strings  of  similar  length  that  resemble  each  other,
often  we  will  be  presented  with  strings  that  are  mostly  dissimilar  except  for  some
unknown  region  of  the  strings,  which  may  represent  a  shared  gene.  To  find  such
genes,  we  need  to  modify  global  alignment  to  instead  search  for  shared  motifs  in  the  form  of  locally
similar  regions  (recall  “Finding  a  Shared  Motif”  and  “Finding  a  Shared  Spliced  Motif”).
 Using  global  alignment  often  fails  to  find  shared  motifs  hidden  in  larger  strings  because  (especially
if  the  similar  region  is  found  on  different  ends  of  the  string)  aligning  the  strings  causes  gap  penalties
to  rack  up.

If  we  are  only  interested  in  comparing  the  regions  of  similarity,  then  we  would  like  to  have  some
way  of  disregarding  the  parts  of  the  strings  that  don't  resemble  each  other.  The  way  to  do  this  is  to
produce  alignment  scores  for  all  possible  pairs  of  substrings.

Problem

A  local  alignment  of  two  strings    and    is  an  alignment  of  substrings    and    of    and   ,
respectively.  Let    denote  the  score  of  an  optimal  alignment  of    and    with  respect  to  some
predetermined  alignment  score.

Given:  Two  protein  strings    and    in  FASTA  format  (each  having  length  at  most  1000  aa).
Return:  A  maximum  alignment  score  along  with  substrings    and    of    and   ,  respectively,  which
produce  this  maximum  alignment  score  (multiple  solutions  may  exist,  in  which  case  you  may  output
any  one).  Use:

The  PAM250  scoring  matrix.

Linear  gap  penalty  equal  to  5.

Sample  Dataset

>Rosalind_80
MEANLYPRTEINSTRING
>Rosalind_21
PLEASANTLYEINSTEIN

Sample  Output

23
LYPRTEINSTRIN
LYEINSTEIN

Problem  87
Inferring  Genotype  from  a  Pedigree

Lying  in  Wait

Single  gene  disorders  can  be  encoded  by  either  dominant  or  recessive  alleles.  In
the  latter  case,  the  affected  person  usually  has  two  healthy  carrier  parents,  who
were  usually  unaware  that  their  child  could  inherit  a  deadly  or  debilitating  genetic
condition  from  them.
 We  know  from  Mendel's  first  law  that  any  offspring  of  two  heterozygous  carriers  has  a  25%  chance
of  inheriting  a  recessive  disorder.  Knowing  your  own  genotype  is  therefore  important  when  deciding
to  have  children,  and  genetic  screening  will  prove  vital  for  preventive  medicine  in  the  coming  years.

In  this  problem,  we  will  consider  an  exercise  in  which  we  determine  the  probability  of  an  organism
exhibiting  each  possible  genotype  for  a  factor  knowing  only  the  genotypes  of  the  organism's
ancestors.

Problem

A  rooted  binary  tree  can  be  used  to  model  the  pedigree  of  an
individual.  In  this  case,  rather  than  time  progressing  from  the
root  to  the  leaves,  the  tree  is  viewed  upside  down  with  time
progressing  from  an  individual's  ancestors  (at  the  leaves)  to
the  individual  (at  the  root).

An  example  of  a  pedigree  for  a  single  factor  in  which  only  the
genotypes  of  ancestors  are  given  is  shown  in  Figure  1.

Given:  A  rooted  binary  tree    in  Newick  format  encoding

an  individual's  pedigree  for  a  Mendelian  factor  whose  alleles
are  A  (dominant)  and  a  (recessive).

Return:  Three  numbers  between  0  and  1,  corresponding  to Figure  1.  The  rooted  binary  tree
the  respective  probabilities  that  the  individual  at  the  root  of   whose  Newick  format  is  (aa,
will  exhibit  the  "AA",  "Aa"  and  "aa"  genotypes. ((Aa,AA),AA)).  Each  leaf  encodes  the
genotype  of  an  ancestor  for  the  given
individual,  which  is  represented  by
Sample  Dataset '?'.

((((Aa,aa),(Aa,Aa)),((aa,aa),(aa,AA))),Aa);

Sample  Output

0.156  0.5  0.344

Problem  88
Maximizing  the  Gap  Symbols  of  an  Optimal  Alignment

Adjusting  Alignment  Parameters

As  we  change  the  parameters  contributing  to  alignment  score,  the  nature  of
alignments  achieving  the  maximum  score  may  change.  One  feature  of  maximum-­
score  alignments  worthy  of  consideration  is  the  number  and  size  of  their  gaps.  In
this  problem,  we  would  like  to  determine  the  maximum  number  of  possible  gaps
in  any  optimal  alignment  based  solely  on  the  parameter  values  chosen.
 Problem

For  the  computation  of  an  alignment  score  generalizing  the  edit  alignment  score,  let    denote  the
score  assigned  to  matched  symbols,    denote  the  score  assigned  to  mismatched  non-­gap  symbols,
and    denote  the  score  assigned  a  symbol  matched  to  a  gap  symbol  '-­'  (i.e.,    is  a  linear  gap
penalty).

Given:  Two  DNA  strings    and    in  FASTA  format  (each  of  length  at  most  5000  bp).
Return:  The  maximum  number  of  gap  symbols  that  can  appear  in  any  maximum  score  alignment  of
 and    with  score  parameters  satisfying   ,   ,  and   .

Sample  Dataset

>Rosalind_92
AACGTA
>Rosalind_47
ACACCTA

Sample  Output

Problem  89
Identifying  Maximal  Repeats

Spies  in  the  War  Against  Phages

In  “Locating  Restriction  Sites”,  we  saw

how  one  weapon  used  by  bacteria  in  their
age-­old  fight  with  phages  is  the  use  of
restriction  enzymes.  Another  defense
Figure  1.  A  genomic  region
mechanism  found  in  the  genomes  of  most  bacteria  and containing  a  CRISPR.  Red
archaea  centers  on  intervals  of  DNA  called  CRISPRs substrings  correspond  to  CRISPR
repeats,  and  blue  substrings
(Clustered  Regularly  Interspaced  Short  Palindromic correspond  to  unique  spacers.
Repeats),  which  allow  the  cell  to  distinguish  its  own  DNA Repeats  are  highly  palindromic  and
from  that  of  phages  or  plasmids. fold  into  a  hairpin  loop  when
transcribed.
Specifically,  a  CRISPR  is  an  interval  of  DNA  consisting  of
identical  repeats  (approximately  23  to  47  bp  long),  alternating  with  unique  intervals  (approximately
21  to  72  bp  long)  called  spacers;;  see  Figure  1.  Spacers  correspond  to  fragments  of  foreign  DNA
that  were  integrated  into  the  genome  between  repeats  and  serve  as  a  memory  bank  for  genetic
material  captured  from  invading  phages.  As  a  result,  spacers  can  be  used  to  recognize  and  silence
invasive  elements.

Specifically,  CRISPRs  are  transcribed  into  RNA  molecules,  each  consisting  of  a  spacer  flanked  by
 partial  repeats.  The  small  CRISPR  RNAs,  together  with  associated  proteins  translated  from  this
RNA,  target  foreign  DNA  that  matches  the  CRISPR  spacer.  In  eukaryotes,  a  similar  process  is
achieved  by  a  process  called  RNA  interference  (RNAi).

To  locate  a  CRISPR  in  a  genome,  we  need  to  search  for  its  repeats.  We  have  already  located  long
repeats  in  “Finding  the  Longest  Multiple  Repeat”,  but  the  case  here  is  different  because  of  the
repeats  appearing  in  CRISPRS  are  relatively  short.  Instead,  we  are  looking  for  repeated  intervals
that  cannot  be  lengthened  in  either  direction  (otherwise,  we  would  intersect  with  a  spacer).

Problem

A  maximal  repeat  of  a  string    is  a  repeated  substring    of    having  two  occurrences    and    such
that    and    cannot  be  extended  by  one  symbol  in  either  direction  in    and  still  agree.

For  example,  "AG"  is  a  maximal  repeat  in  "TAGTTAGCGAGA"  because  even  though  the  first  two
occurrences  of  "AG"  can  be  extended  left  into  "TAG",  the  first  and  third  occurrences  differ  on  both
sides  of  the  repeat;;  thus,  we  conclude  that  "AG"  is  a  maximal  repeat.  Note  that  "TAG"  is  also  a
maximal  repeat  of  "TAGTTAGCGAGA",  since  its  only  two  occurrences  do  not  still  match  if  we  extend
them  in  either  direction.

Given:  A  DNA  string    of  length  at  most  1  kbp.

Return:  A  list  containing  all  maximal  repeats  of    having  length  at  least  20.

Sample  Dataset

TAGAGATAGAATGGGTCCAGAGTTTTGTAATTTCCATGGGTCCAGAGTTTTGTAATTTATTATATAGAGAT
AGAATGGGTCCAGAGTTTTGTAATTTCCATGGGTCCAGAGTTTTGTAATTTAT

Sample  Output

TAGAGATAGAATGGGTCCAGAGTTTTGTAATTTCCATGGGTCCAGAGTTTTGTAATTTAT
ATGGGTCCAGAGTTTTGTAATTT

Hint

How  can  we  use  the  suffix  tree  of    to  find  maximal  repeats?

Problem  90
Multiple  Alignment

Comparing  Multiple  Strings  Simultaneously

In  “Consensus  and  Profile”,  we  generalized  the  notion  of  Hamming  distance  to  find  an  average  case
 for  a  collection  of  nucleic  acids  or  peptides.  However,  this  method  only  worked  if
the  polymers  had  the  same  length.  As  we  have  already  noted  in  “Edit  Distance”,
homologous  strands  of  DNA  have  varying  lengths  because  of  the  effect  of
mutations  inserting  and  deleting  intervals  of  genetic  material;;  as  a  result,  we  need
to  generalize  the  notion  of  alignment  to  cover  multiple  strings.

Problem

A  multiple  alignment  of  a  collection  of  three  or  more  strings  is  formed  by  adding  gap  symbols  to  the
strings  to  produce  a  collection  of  augmented  strings  all  having  the  same  length.

A  multiple  alignment  score  is  obtained  by  taking  the  sum  of  an  alignment  score  over  all  possible
pairs  of  augmented  strings.  The  only  difference  in  scoring  the  alignment  of  two  strings  is  that  two  gap
symbols  may  be  aligned  for  a  given  pair  (requiring  us  to  specify  a  score  for  matched  gap  symbols).

Given:  A  collection  of  four  DNA  strings  of  length  at  most  10  bp  in  FASTA  format.
Return:  A  multiple  alignment  of  the  strings  having  maximum  score,  where  we  score  matched
symbols  0  (including  matched  gap  symbols)  and  all  mismatched  symbols  -­1  (thus  incorporating  a
linear  gap  penalty  of  1).

Sample  Dataset

>Rosalind_7
ATATCCG
>Rosalind_35
TCCG
>Rosalind_23
ATGTACTG
>Rosalind_44
ATGTCTG

Sample  Output

-­‐18
ATAT-­‐CCG
-­‐T-­‐-­‐-­‐CCG
ATGTACTG
ATGT-­‐CTG

Problem  91
Creating  a  Restriction  Map

Genetic  Fingerprinting
 Recall  that  a  restriction  enzyme  cuts  the  endpoints  of  a  specific  interval  of  DNA,
which  must  form  a  reverse  palindrome  that  typically  has  length  4  or  6.  The  interval
of  DNA  cleaved  by  a  given  restriction  enzyme  is  called  its  recognition  sequence.

A  single  human  chromosome  is  so  long  that  a  given  recognition  sequence  will
occur  frequently  throughout  the  chromosome  (recall  from  “Expected  Number  of
Restriction  Sites”  that  a  recognition  sequence  would  be  expected  to  occur  several
times  even  in  a  short  chromosome).  Nevertheless,  the  small-­scale  mutations  that
create  diversity  in  the  human  genome  (chiefly  SNPs)  will  cause  each  human  to  have  a  different
collection  of  recognition  sequences  for  a  given  restriction  enzyme.

Genetic  fingerprinting  is  the  term  applied  to  the  general  process  of  forming  a  limited  picture  of  a
person's  genetic  makeup  (which  was  traditionally  cheaper  than  sequencing).  The  earliest
application  of  genetic  fingerprinting  inexpensive  enough  to  be  widely  used  in  common  applications,
like  forensics  and  paternity  tests,  relied  on  a  process  called  restriction  digest.  In  this  technique,  a
sample  of  DNA  is  replicated  artificially,  then  treated  with  a  given  restriction  enzyme;;  when  the
enzyme  cuts  the  DNA  at  restriction  sites,  it  forms  a  number  of  fragments.  A  second  process  called
gel  electrophoresis  then  separates  these  fragments  along  a  membrane  based  on  their  size,  with
larger  pieces  tending  toward  one  end  and  smaller  pieces  tending  toward  the  other.  When  the
membrane  is  stained  or  viewed  with  an  X-­ray  machine,  the  fragments  create  a  distinct  banding
pattern,  which  typically  differs  for  any  two  individuals.

These  intervals  can  be  thought  of  simply  as  the  collection  of  distances  between  restriction  sites  in
the  genome.  Before  the  rapid  advances  of  genome  sequencing,  biologists  wanted  to  know  if  they
could  use  only  these  distances  to  reconstruct  the  actual  locations  of  restriction  sites  in  the
genome,  forming  a  restriction  map.  Restriction  maps  were  desired  in  the  years  before  the  advent
of  sequencing,  when  any  information  at  all  about  genomic  makeup  was  highly  coveted.  The
application  of  forming  a  restriction  map  from  cleaved  restriction  fragments  motivates  the  following
problem.

Problem

For  a  set    containing  numbers,  the  difference  multiset  of  

 is  the  multiset    defined  as  the  collection  of  all  positive
differences  between  elements  of   .  As  a  quick  example,  if  
,  then  we  will  have  that   .

If    contains    elements,  then    will  contain  one  element

Figure  1.  In  the  simplified  figure
for  each  pair  of  elements  from   ,  so  that    contains   above,  we  know  that  the  dashed
elements  (see  combination  statistic).  You  may  note  the segments  came  from  a
chromosome;;  we  desire  a  collection
similarity  between  the  difference  multiset  and  the  Minkowski of  numbers  whose  differences
difference   ,  which  contains  the  elements  of    and match  the  lengths  of  the  dotted
lines,  which  will  correspond  to  the
their  negatives.  For  the  above  set   ,    is   locations  of  restriction  sites  on  the
. unknown  chromosome.  Taken  from
Jones  &  Pevzner,  An  Introduction  to
In  practical  terms,  we  can  easily  obtain  a  multiset   Bioinformatics  Algorithms.
corresponding  to  the  distances  between  restriction  sites  on  a
chromosome.  If  we  can  find  a  set    whose  difference  multiset    is  equal  to   ,  then    will
represent  possible  locations  of  these  restriction  sites.  For  an  example,  consult  Figure  1.

Given:  A  multiset    containing    positive  integers  for  some  positive  integer   .

Return:  A  set    containing    nonnegative  integers  such  that   .

Sample  Dataset
 2  2  3  3  4  5  6  7  8  10

Sample  Output

0  2  4  7  10

Problem  92
Counting  Rooted  Binary  Trees

From  Unrooted  to  Rooted  Trees

Recall  that  a  rooted  binary  tree  is  a  binary  tree  for  which  the  root  is  the  only
node  of  degree  2.  Such  a  tree  differs  from  an  unrooted  binary  tree  only  in  the
existence  of  the  root.

Different  phylogenetic  methods  may  be  better  suited  to  rooted  or  unrooted  trees.  If  a  method
produces  an  unrooted  tree,  then  the  root  (i.e.,  the  common  ancestor  of  all  taxa)  could  theoretically
be  placed  anywhere.  Thus,  there  will  be  more  rooted  binary  trees  than  unrooted  binary  trees  on  the
same  number  of  taxa.  The  question  is:  how  many  more  rooted  trees  are  there?

Problem

As  in  the  case  of  unrooted  trees,  say  that  we  have  a  fixed  collection  of    taxa  labeling  the  leaves  of  a
rooted  binary  tree   .  You  may  like  to  verify  that  (by  extension  of  “Counting  Phylogenetic  Ancestors”)
such  a  tree  will  contain    internal  nodes  and    total  edges.  Any  edge  will  still  encode  a
split  of  taxa;;  however,  the  two  splits  corresponding  to  the  edges  incident  to  the  root  of    will  be  equal.
We  still  consider  two  trees  to  be  equivalent  if  they  have  the  same  splits  (which  requires  that  they  must
also  share  the  same  duplicated  split  to  be  equal).

Let    represent  the  total  number  of  distinct  rooted  binary  trees  on    labeled  taxa.

Given:  A  positive  integer    ( ).

Return:  The  value  of    modulo  1,000,000.

Sample  Dataset

Sample  Output

15
Problem  93
Sex-­Linked  Inheritance

Chromosomes  Determine  Sex

In  “Independent  Segregation  of

Chromosomes”,  we  discussed  how
chromosomes  in  diploid  organisms  form
pairs  of  homologs.  It  turns  out  that  this  is
not  the  case  for  one  pair  of  chromosomes  in  animals.  In
1905,  Nettie  Stevens  and  Edmund  Wilson  independently
discovered  that  male  animals  possess  one  chromosome
that  is  shorter  than  its  partner,  whereas  female  animals
instead  have  two  long  chromosomes.  The  shorter
chromosome  earned  the  title  of  Y  chromosome  for  its
stunted  shape,  whereas  the  longer  chromosome  became
known  as  the  X  chromosome.  These  two  chromosomes
are  aptly  termed  sex  chromosomes,  or  allosomes,  and Figure  1.  Morgan's  two  experiments
on  fruit  fly  eye  color.  In  the  first
we  write  the  female  sex  chromosome  genotype  as   experiment,  a  white-­eyed  male  is
and  the  male  genotype  as   .  The  remaining crossed  with  a  purebred  red-­eyed
homologous  chromosome  pairs  are  called  autosomes. female;;  in  the  second  experiment,  a
red-­eyed  male  is  crossed  with  a
white-­eyed  female.  The  results  of
Sex  chromosomes  are  still  passed  on  to  gametes  based
Morgan's  expermients  demonstrate
on  the  outcome  of  a  coin  flip,  but  egg  cells  (deriving  from that  eye  color  must  be  encoded  by  a
females)  must  always  possess  an  X  chromosome,  so  that recessive  allele  on  the  X
chromosome.
the  sex  of  an  individual  is  determined  by  whether  it
receives  an  X  or  a  Y  chromosome  from  its  father's  sperm
cell.

Fast-­forward  five  years  to  1910  and  the  lab  of  Thomas  Hunt  Morgan,  who  is  often  considered  the
first  modern  geneticist  because  of  his  tireless  work  to  place  Mendel's  work  on  sound  footing.  One
of  Morgan's  many  experiments  with  fruit  flies  (genus  Drosophila)  began  as  he  noticed  a  number  of
white-­eyed  males.  When  these  white-­eyed  flies  were  crossed  with  purebred  red-­eyed  females,  their
progeny  were  all  red-­eyed,  and  yet  crossing  the  second  generation's  red-­eyed  individuals  with  each
other  produced  some  white-­eyed  males  but  exclusively  red-­eyed  females.  Strange  results  indeed.

Morgan's  experiments  are  summarized  in  Figure  1,  after  which  he  concluded  that  the  trait  for  eye
color  in  fruit  flies  must  be  sex  linked,  or  encoded  on  a  sex  chromosome.  More  specifically,  the
factor  for  white  eye  color  is  encoded  by  a  recessive  allele  on  the  X  chromosome.  Because  a  male
only  has  one  copy  of  the  X  chromosome,  having  only  one  recessive  allele  will  cause  the  individual
to  exhibit  white  eyes,  whereas  a  female  fly  requires  both  copies  of  the  recessive  allele  to  possess
white  eyes.

X-­linked  recessive  traits  are  manifested  in  males  much  more  often  than  in  females,  because  a  male
only  needs  to  receive  a  recessive  allele  from  his  mother  to  exhibit  the  trait:  in  the  case  of  genetic
conditions,  half  of  all  male  children  born  to  carrier  mothers  will  inherit  the  condition.

Problem
 The  conditional  probability  of  an  event    given  another  event   ,  written   ,  is  equal  to  
 divided  by   .

Note  that  if    and    are  independent,  then    must  be  equal  to   ,  which
results  in   .  This  equation  offers  an  intuitive  view  of  independence:  the  probability
of   ,  given  the  occurrence  of  event   ,  is  simply  the  probability  of    (which  does  not  depend  on   ).

In  the  context  of  sex-­linked  traits,  genetic  equilibrium  requires  that  the  alleles  for  a  gene    are
uniformly  distributed  over  the  males  and  females  of  a  population.  In  other  words,  the  distribution  of
alleles  is  independent  of  sex.

Given:  An  array    of  length    for  which    represents  the  proportion  of  males  in  a  population
exhibiting  the   -­th  of    total  recessive  X-­linked  genes.  Assume  that  the  population  is  in  genetic
equilibrium  for  all    genes.

Return:  An  array    of  length    in  which    equals  the  probability  that  a  randomly  selected
female  will  be  a  carrier  for  the   -­th  gene.

Sample  Dataset

0.1  0.5  0.8

Sample  Output

0.18  0.5  0.32

Problem  94
Phylogeny  Comparison  with  Split  Distance

Quantifying  Binary  Tree  Comparison

We  may  often  obtain  two  different  phylogenies  on  the  same  collection  of  taxa  from
different  sets  of  data.  As  a  result,  we  would  like  to  have  a  way  of  quantifying  how
much  the  two  phylogenies  differ.  In  the  simplest  case,  we  would  like  to  compare
the  characters  of  two  phylogenies.

Recall  from  “Counting  Unrooted  Binary  Trees”  that  two  unrooted  binary  trees  are  equivalent  when
they  have  the  same  set  of  splits;;  recall  also  (by  extension  of  “Counting  Phylogenetic  Ancestors”)
that  any  unrooted  binary  tree  on    taxa  must  have    nontrivial  splits.

Problem

Define  the  split  distance  between  two  unrooted  binary  trees  as  the  number  of  nontrivial  splits
contained  in  one  tree  but  not  the  other.

Formally,  if    denotes  the  number  of  nontrivial  splits  shared  by  unrooted  binary  trees    and  
,  Then  their  split  distance  is   .
 Given:  A  collection  of  at  most  3,000  species  taxa  and  two  unrooted  binary  trees    and    on
these  taxa  in  Newick  format.

Return:  The  split  distance   .

Sample  Dataset

dog  rat  elephant  mouse  cat  rabbit

(rat,(dog,cat),(rabbit,(elephant,mouse)));
(rat,(cat,dog),(elephant,(mouse,rabbit)));

Sample  Output

Problem  95
The  Wright-­Fisher  Model  of  Genetic  Drift

Hardy-­Weinberg  Revisited

The  principle  of  genetic  equilibrium  is  an  idealistic  model  for  population  genetics
that  simply  cannot  hold  for  all  genes  in  practice.  For  one,  evolution  has  proven  too
powerful  for  equilibrium  to  possibly  hold.  At  the  same  time,  evolution  works  on  the
scale  of  eons,  and  at  any  given  moment  in  time,  most  populations  are  essentially
stable.

Yet  we  could  overlook  the  inevitable  effects  of  simple  random  chance  in  disrupting  the  allelic
frequency  for  a  given  gene,  a  phenomenon  called  genetic  drift.

In  this  problem,  we  would  like  to  obtain  a  simple  mathematical  model  of  genetic  drift,  and  so  we  will
need  to  make  a  number  of  simplifying  assumptions.  First,  assume  that  individuals  from  different
generations  do  not  mate  with  each  other,  so  that  generations  exist  as  discrete,  non-­overlapping
quantities.  Second,  rather  than  selecting  pairs  of  mating  organisms,  we  simply  randomly  select  the
alleles  for  the  individuals  of  the  next  generation  based  on  the  allelic  frequency  in  the  present
generation.  Third,  the  population  size  is  stable,  so  that  we  do  not  need  to  take  into  account  the
population  growing  or  shrinking  between  generations.  Taken  together,  these  three  assumptions
make  up  the  Wright-­Fisher  model  of  genetic  drift.

Problem

Consider  flipping  a  weighted  coin  that  gives  "heads"  with  some  fixed  probability    (i.e.,    is  not
necessarily  equal  to  1/2).

We  generalize  the  notion  of  binomial  random  variable  from  “Independent  Segregation  of  Chromosomes”
to  quantify  the  sum  of  the  weighted  coin  flips.  Such  a  random  variable    takes  a  value  of    if  a
sequence  of    independent  "weighted  coin  flips"  yields    "heads"  and    "tails."  We  write  that  
 .

To  quantify  the  Wright-­Fisher  Model  of  genetic  drift,  consider  a  population  of    diploid  individuals,
whose    chromosomes  possess    copies  of  the  dominant  allele.  As  in  “Counting  Disease
Carriers”,  set   .  Next,  recall  that  the  next  generation  must  contain  exactly    individuals.
These  individuals'    alleles  are  selected  independently:  a  dominant  allele  is  chosen  with  probability  
,  and  a  recessive  allele  is  chosen  with  probability   .

Given:  Positive  integers    ( ),    ( ),    ( )  and    ( ).

Return:  The  probability  that  in  a  population  of    diploid  individuals  initially  possessing    copies  of
a  dominant  allele,  we  will  observe  after    generations  at  least    copies  of  a  recessive  allele.  Assume
the  Wright-­Fisher  model.

Sample  Dataset

4  6  2  1

Sample  Output

0.772

Problem  96
Alignment-­Based  Phylogeny

From  Characters  Toward  Alignments

In  “Creating  a  Character  Table  from  Genetic  Strings”,  we  used  strings  to  create  a
collection  of  characters  from  which  we  could  create  a  phylogeny.  However,  the
strings  all  had  to  share  the  same  length,  which  was  a  problem.  In  practice,  we
would  like  to  create  a  phylogeny  from  genetic  strings  having  differing  lengths;;
specifically,  our  aim  is  to  construct  a  phylogeny  from  a  multiple  alignment.

Unfortunately,  constructing  a  phylogeny  from  the  ground  up  based  only  on  an  alignment  can  be
difficult.  In  order  to  produce  an  efficient  solution,  we  will  need  to  assume  that  the  structure  of  the
phylogeny  has  already  been  provided  (perhaps  from  character-­based  methods),  and  our  aim  instead
is  to  reconstruct  the  genetic  strings  corresponding  to  the  internal  nodes  (i.e.,  ancestors)  in  the  tree.

The  ancestor  strings  should  have  the  property  that  the  total  number  of  point  mutations  separating
adjacent  nodes  in  the  tree  is  minimized  (in  keeping  with  parsimony).

Problem

Say  that  we  have    taxa  represented  by  strings    with  a  multiple  alignment  inducing
corresponding  augmented  strings   .

Recall  that  the  number  of  single-­symbol  substitutions  required  to  transform  one  string  into  another  is
the  Hamming  distance  between  the  strings  (see  “Counting  Point  Mutations”).  Say  that  we  have  a
rooted  binary  tree    containing    at  its  leaves  and  additional  strings  
 at  its  internal  nodes,  including  the  root  (the  number  of  internal  nodes  is  
 by  extension  of  “Counting  Phylogenetic  Ancestors”).  Define    as  the  sum  of  
over  all  edges    in   :

Thus,  our  aim  is  to  minimize   .

Given:  A  rooted  binary  tree    on    ( )  species,  given  in  Newick  format,  followed  by  a
multiple  alignment  of    ( )  augmented  DNA  strings  having  the  same  length  (at  most  300  bp)
corresponding  to  the  species  and  given  in  FASTA  format.

Return:  The  minimum  possible  value  of   ,  followed  by  a  collection  of  DNA  strings  to  be
assigned  to  the  internal  nodes  of    that  will  minimize    (multiple  solutions  will  exist,  but  you
need  only  output  one).

Sample  Dataset

(((ostrich,cat)rat,(duck,fly)mouse)dog,(elephant,pikachu)hamster)robot;
>ostrich
AC
>cat
CA
>duck
T-­‐
>fly
GC
>elephant
-­‐T
>pikachu
AA

Sample  Output

8
>rat
AC
>mouse
TC
>dog
AC
>hamster
AT
>robot
AC

Note

Given  internal  strings  minimizing   ,  the  alignment  between  any  two  adjacent  strings  is  not
 necessarily  an  optimal  global  paired  alignment.  In  other  words,  it  may  not  be  the  case  that  
 is  equal  to  the  edit  distance   .

Problem  97
Assessing  Assembly  Quality  with  N50  and  N75

How  Well  Assembled  Are  Our  Contigs?

As  we  have  stated,  the  goal  of  genome  sequencing  is  to  create  contigs  that  are
as  long  as  possible.  Thus,  after  fragment  assembly,  it  is  important  to  possess
statistics  quantifying  how  well-­assembled  our  contigs  are.

First  and  foremost,  we  demand  a  measure  of  what  percentage  of  the  assembled  genome  is  made
up  of  long  contigs.  Our  first  question  is  then:  if  we  select  contigs  from  our  collection,  how  long  do
the  contigs  need  to  be  to  cover  50%  of  the  genome?

Problem

Given  a  collection  of  DNA  strings  representing  contigs,  we  use  the  N  statistic  NXX  (where  XX  ranges
from  01  to  99)  to  represent  the  maximum  positive  integer    such  that  the  total  number  of  nucleotides
of  all  contigs  having  length    is  at  least  XX%  of  the  sum  of  contig  lengths.  The  most  commonly
used  such  statistic  is  N50,  although  N75  is  also  worth  mentioning.

Given:  A  collection  of  at  most  1000  DNA  strings  (whose  combined  length  does  not  exceed  50  kbp).
Return:  N50  and  N75  for  this  collection  of  strings.

Sample  Dataset

GATTACA
TACTACTAC
ATTGAT
GAAGA

Sample  Output

7  6

Extra  Information

For  an  explanation  of  the  results  obtained  in  the  sample  above,  contigs  of  length  at  least  7  total  7  +
9  =  16  bp,  which  is  more  than  50%  of  the  total  27).  Contigs  of  length  at  least  8  total  only  9  bp  (less
than  50%).
 Contigs  of  length  at  least  6  total  6  +  7  +  9  =  22  bp,  which  is  more  than  75%  of  all  base  pairs.
Contigs  of  length  at  least  7  total  only  16  bp  (less  than  75%).

Problem  98
Fixing  an  Inconsistent  Character  Set

Pitfalls  of  Character-­Based  Phylogeny

In  “Character-­Based  Phylogeny”,  we  asked  for  the  construction  of  an  unrooted
binary  tree  from  a  consistent  character  table.  However,  the  assumption  of
consistency  is  often  inaccurate,  as  many  character  collections  derived  from  real
data  are  inconsistent,  owing  to  the  fact  that  the  reinforcement  of  mutations  by
evolution  can  cause  species  to  lose  features  over  time,  evolve  to  produce  the  same  character  on
different  evolutionary  paths,  or  revert  to  a  past  character.  This  issue  arises  even  when  using
genetic  characters  taken  from  SNPs,  as  point  mutations  can  be  undone.

As  an  example  of  why  using  characters  can  lead  us  astray,  let's  return  to  our  first  example  of  a
character  introduced  in  “Character-­Based  Phylogeny”.  There,  we  learned  that  dinosaurs  may  be
divided  into  the  two  Orders  Saurischia  and  Ornithischia  depending  on  hip-­bone  shape:  the  former
have  "lizard  hips,"  whereas  the  latter  have  "bird  hips."  Adding  to  this  information  the  fact  that  birds
are  widely  believed  to  descend  from  dinosaurs,  we  would  guess  that  birds  derive  from
ornithischians.  Yet  this  is  not  the  case:  birds  derive  from  theropods,  a  suborder  of  the  saurischians!
The  shared  hip  bone  shape  with  ornithischians  is  either  simply  coincidence  or  caused  by  the
"convergence"  of  bird  hip  shape  with  that  of  ornithischians  along  their  different  evolutionary  paths.

Another  example  of  a  character  that  would  be  ill-­suited  for  phylogenetic  analysis  is  the  presence  or
absence  of  wings  in  insects.  Many  wingless  modern  species  wingless  modern  species  have
evolved  from  wildly  differing  ancestors  that  lost  their  wings  independently  of  each  other.

If  we  divided  a  collection  of  taxa  based  on  either  of  these  characters,  many  different  taxa  would  be
lumped  together  when  they  do  not  in  fact  share  a  recent  common  ancestor,  which  could  have
disastrous  consequences  when  trying  to  assign  characters  to  the  splits  of  a  phylogeny.

The  moral  is  that  we  must  select  our  characters  carefully,  although  the  Catch-­22  is  that  we  don't
know  in  advance  which  characters  are  the  most  appropriate  to  use  until  we  actually  start
constructing  phylogenies.  At  the  same  time,  if  we  err  on  the  side  of  caution,  then  using  too  few
characters  might  not  provide  us  with  enough  splits  to  generate  an  unrooted  binary  tree,  thus
inducing  an  enormous  number  of  possible  phylogenies  (recall  “Counting  Unrooted  Binary  Trees”  and
how  quickly  the  total  number  of  trees  grows  with  the  number  of  taxa).

Problem

A  submatrix  of  a  matrix    is  a  matrix  formed  by  selecting  rows  and  columns  from    and  taking
only  those  entries  found  at  the  intersections  of  the  selected  rows  and  columns.  We  may  also  think  of  a
submatrix  as  formed  by  deleting  the  remaining  rows  and  columns  from   .

Given:  An  inconsistent  character  table    on  at  most  100  taxa.
Return:  A  submatrix  of    representing  a  consistent  character  table  on  the  same  taxa  and  formed
 by  deleting  a  single  row  of   .  (If  multiple  solutions  exist,  you  may  return  any  one.)

Sample  Dataset

100001
000110
111000
100111

Sample  Output

000110
100001
100111

Problem  99
Wright-­Fisher's  Expected  Behavior

Reaching  Population  Equilibrium

In  “The  Wright-­Fisher  Model  of  Genetic  Drift”,  we  introduced  the  Wright-­Fisher
model  of  genetic  drift.  Although  the  effects  of  genetic  drift  are  inevitable,  we  should
be  able  to  quantify  how  many  alleles  for  a  given  trait  will  remain  in  the  next
generation.

Intuitively,  because  Wright-­Fisher  demands  that  we  randomly  and  independently  select  alleles  for
the  next  generation  based  off  the  allele  frequency  of  the  present  generation,  we  would  hope  that  on
average  this  frequency  would  illustrate  a  stabilizing  effect:  that  is,  the  expected  frequency  in  the
next  generation  should  equal  the  allele  frequency  in  the  current  generation.  In  this  problem,  we  will
see  if  the  mathematics  matches  our  intuition.

Problem

In  “The  Wright-­Fisher  Model  of  Genetic  Drift”,  we  generalized  the  concept  of  a  binomial  random  variable
 as  a  "weighted  coin  flip."  It  is  only  natural  to  calculate  the  expected  value  of  such  a  random
variable.

For  example,  in  the  case  of  unweighted  coin  flips  (i.e.,   ),  our  intuition  would  indicate  that  
 is   ;;  what  should  be  the  expected  value  of  a  binomial  random  variable?

Given:  A  positive  integer    ( )  followed  by  an  array    of  length    ( )

containing  numbers  between    and   .  Each  element  of    can  be  seen  as  representing  a  probability
corresponding  to  an  allele  frequency.

Return:  An  array    of  length    for  which    is  the  expected  value  of   ;;  in  terms  of
Wright-­Fisher,  it  represents  the  expected  allele  frequency  of  the  next  generation.
 Sample  Dataset

17
0.1  0.2  0.3

Sample  Output

1.7  3.4  5.1

Problem  100
The  Founder  Effect  and  Genetic  Drift

Strength  in  Numbers

Charles  Darwin  is  known  first  and  foremost  for  his  notion  of  natural  selection,
the  elegant  statistical  fact  that  changes  in  populations  are  attributable  to  the
observation  that  organisms  better  equipped  to  handle  their  environment  are  more
likely  to  survive  and  reproduce,  thus  passing  on  their  beneficial  traits  to  the  next
generation.  As  a  result  of  natural  selection,  populations  can  change  greatly  over  a  long  time.

A  lesser  known  aspect  of  Darwin's  evolutionary  theory  dictates  how  new  species  are  actually
created.  Darwin  noted  that  the  only  way  for  a  population  to  grow  so  distinct  that  it  would  actually
split  off  and  form  a  new  species  would  be  if  the  population  were  isolated  for  a  very  long  period.  This
notion  that  isolation  forms  new  species  was  validated  by  Darwin's  observation  that  the  tiny
Galapagos  islands  in  the  South  Pacific  enjoy  a  diversity  of  species  rivaling  that  of  a  much  larger
ecosystem.

Isolated  populations  also  tend  to  be  small,  strengthening  the  effects  of  genetic  drift.  To  take  an
extreme  example,  consider  a  population  of  only  2  organisms  that  are  both  heterozygous  for  a  given
factor.  Note  that  there  is  a  1/8  chance  that  2  offspring  of  these  organisms  will  possess  only
recessive  alleles  or  only  dominant  alleles  for  the  factor,  thus  wiping  out  the  other  allele  completely.

In  general,  the  principle  stating  that  mutations  (both  positive  and  negative)  can  randomly  attain
higher  proportions  in  small,  isolated  communities  than  they  would  in  large  populations,  is  known  as
the  founder  effect.  An  infamous  example  of  the  founder  effect  on  human  populations  occurs  in
Pennsylvania,  where  the  Amish  community  is  at  risk  for  a  much  greater  incidence  of  Ellis-­van
Creveld  syndrome,  a  single  gene  disorder  causing  a  slew  of  defects,  including  additional  fingers
and  toes  (polydactyly).  The  condition  has  been  traced  to  a  single  couple  in  the  original  Amish
settlers,  and  it  is  still  preserved  in  elevated  percentages  because  of  the  community's  isolationism.

In  this  problem,  we  would  like  to  apply  the  Wright-­Fisher  model  of  genetic  drift  to  understand  the
power  of  the  founder  effect.  Specifically,  we  will  quantify  the  likelihood  that  an  allele  will  be
completely  annihilated  in  a  small  population  after  a  number  of  generations.

Problem
 Given:  Two  positive  integers    and   ,  followed  by  an  array    containing    integers  between  0
and   .    represents  the  number  of  recessive  alleles  for  the   -­th  factor  in  a  population  of  
diploid  individuals.

Return:  An    matrix    for  which    represents  the  common  logarithm  of  the  probability
that  after    generations,  no  copies  of  the  recessive  allele  for  the   -­th  factor  will  remain  in  the
population.  Apply  the  Wright-­Fisher  model.

Sample  Dataset

4  3
0  1  2

Sample  Output

0.0  -­‐0.463935575821  -­‐0.999509892866

0.0  -­‐0.301424998891  -­‐0.641668367342
0.0  -­‐0.229066698008  -­‐0.485798552456

Problem  101
Global  Alignment  with  Scoring  Matrix  and  Affine  Gap
Penalty

Mind  the  Gap

In  “Global  Alignment  with  Scoring  Matrix”,  we  considered  a  linear  gap  penalty,  in
which  each  inserted/deleted  symbol  contributes  the  exact  same  amount  to  the
calculation  of  alignment  score.  However,  as  we  mentioned  in  “Global  Alignment
with  Constant  Gap  Penalty”,  a  single  large  insertion/deletion  (due  to  a
rearrangement  is  then  punished  very  strictly,  and  so  we  proposed  a  constant  gap  penalty.

Yet  large  insertions  occur  far  more  rarely  than  small  insertions  and  deletions.  As  a  result,  a  more
practical  method  of  penalizing  gaps  is  to  use  a  hybrid  of  these  two  types  of  penalties  in  which  we
charge  one  constant  penalty  for  beginning  a  gap  and  another  constant  penalty  for  every  additional
symbol  added  or  deleted.

Problem

An  affine  gap  penalty  is  written  as   ,  where    is  the  length  of  the  gap,    is  a  positive
constant  called  the  gap  opening  penalty,  and    is  a  positive  constant  called  the  gap  extension
penalty.

We  can  view  the  gap  opening  penalty  as  charging  for  the  first  gap  symbol,  and  the  gap  extension
penalty  as  charging  for  each  subsequent  symbol  added  to  the  gap.
 For  example,  if    and   ,  then  a  gap  of  length  1  would  be  penalized  by  11  (for  an  average
cost  of  11  per  gap  symbol),  whereas  a  gap  of  length  100  would  have  a  score  of  110  (for  an  average
cost  of  1.10  per  gap  symbol).

Consider  the  strings  "PRTEINS"  and  "PRTWPSEIN".  If  we  use  the  BLOSUM62  scoring  matrix  and  an
affine  gap  penalty  with    and   ,  then  we  obtain  the  following  optimal  alignment.

 PRT-­‐-­‐-­‐EINS
 |||      |||
 PRTWPSEIN-­‐

Matched  symbols  contribute  a  total  of  32  to  the  calculation  of  the  alignment's  score,  and  the  gaps  cost
13  and  11  respectively,  yielding  a  total  score  of  8.

Given:  Two  protein  strings    and    in  FASTA  format  (each  of  length  at  most  100  aa).
Return:  The  maximum  alignment  score  between    and   ,  followed  by  two  augmented  strings  
and    representing  an  optimal  alignment  of    and   .  Use:

The  BLOSUM62  scoring  matrix.

Gap  opening  penalty  equal  to  11.
Gap  extension  penalty  equal  to  1.

Sample  Dataset

>Rosalind_49
PRTEINS
>Rosalind_47
PRTWPSEIN

Sample  Output

8
PRT-­‐-­‐-­‐EINS
PRTWPSEIN-­‐

Problem  102
Genome  Assembly  with  Perfect  Coverage  and  Repeats

Repeats:  A  Practical  Assembly  Difficulty

Genome  assembly  is  straightforward  if  we  know  in  advance  that  the  de  Bruijn
graph  has  exactly  one  directed  cycle  (see  “Genome  Assembly  with  Perfect
Coverage”).

In  practice,  a  genome  contains  repeats  longer  than  the  length  of  the  k-­mers  that  we  wish  to  use  to
assemble  the  genome.  Such  repeats  increase  the  number  of  cycles  present  in  the  de  Bruijn  graph
 for  these   -­mers,  thus  preventing  us  from  assembling  the  genome  uniquely.

For  example,  consider  the  circular  string  (ACCTCCGCC),  along  with  a  collection    of  error-­free
reads  of  length  3,  exhibiting  perfect  coverage  and  taken  from  the  same  strand  of  an  interval  of  DNA.
The  corresponding  de  Bruijn  graph    (where  edges  correspond  to  3-­mers  and  nodes  correspond
to  2-­mers)  has  at  least  two  directed  cycles:  one  giving  the  original  circular  string  (ACCTCCGCC),
and  another  corresponding  to  the  misfit  (ACCGCCTCC).

Also,  note  that  these  cycles  are  not  simple  cycles,  as  the  node  corresponding  to  "CC"  is  visited
three  times  in  each  cycle.

To  generalize  the  problem  of  genome  assembly  from  a  de  Bruijn  graph  to  the  case  of  genomes
containing  repeats,  we  therefore  must  add  a  constraint:  in  a  cycle  corresponding  to  a  valid
assembly,  every  3-­mer  must  appear  as  many  times  in  the  cycle  as  it  does  in  our  collection  of
reads  (which  correspond  to  all  3-­mers  in  the  original  string).

Problem

Recall  that  a  directed  cycle  is  a  cycle  in  a  directed  graph  in  which  the  head  of  one  edge  is  equal  to  the
tail  of  the  following  edge.

In  a  de  Bruijn  graph  of   -­mers,  a  circular  string    is  constructed  from  a  directed  cycle  
 is  given  by   .  That  is,  because
the  final    symbols  of    overlap  with  the  first    symbols  of   ,  we  simply  tack  on  the   -­th
symbol  of    to   ,  then  iterate  the  process.

For  example,  the  circular  string  assembled  from  the  cycle  "AC"    "CT"    "TA"    "AC"  is  simply
(ACT).  Note  that  this  string  only  has  length  three  because  the  2-­mers  "wrap  around"  in  the  string.

If  every   -­mer  in  a  collection  of  reads  occurs  as  an  edge  in  a  de  Bruijn  graph  cycle  the  same  number
of  times  as  it  appears  in  the  reads,  then  we  say  that  the  cycle  is  "complete."

Given:  A  list    of  error-­free  DNA   -­mers  ( )  taken  from  the  same  strand  of  a
circular  chromosome  (of  length   ).

Return:  All  circular  strings  assembled  by  complete  cycles  in  the  de  Bruijn  graph    of   .  The
strings  may  be  given  in  any  order,  but  each  one  should  begin  with  the  first   -­mer  provided  in  the
input.

Sample  Dataset

CAG
AGT
GTT
TTT
TTG
TGG
GGC
GCG
CGT
GTT
TTC
TCA
CAA
AAT
ATT
 TTC
TCA

Sample  Output

CAGTTCAATTTGGCGTT
CAGTTCAATTGGCGTTT
CAGTTTCAATTGGCGTT
CAGTTTGGCGTTCAATT
CAGTTGGCGTTCAATTT
CAGTTGGCGTTTCAATT

Problem  103
Overlap  Alignment

Overlapping  Reads  with  Errors

As  also  mentioned  in  “Error  Correction  in  Reads”,  the  sequencing  machines  that
identify  reads  can  make  errors.  However,  the  problem  that  we  considered  in
“Genome  Assembly  as  Shortest  Superstring”  assumed  that  all  reads  are  error-­
free.

Thus,  rather  than  trying  to  overlap  reads  exactly,  we  will  instead  do  so  approximately.  The  key  to
do  this  is  to  move  toward  methods  that  incorporate  alignments.  Yet  neither  global  nor  local
alignment  is  appropriate  for  this  task.  Global  alignment  will  attempt  to  align  the  entire  reads,  when
we  know  that  only  the  overlapping  parts  of  the  reads  are  relevant.  For  that  matter,  we  may  identify
an  optimal  local  alignment  that  does  not  correspond  to  an  overlap.

As  a  result,  we  need  a  specific  type  of  local  alignment  that  aligns  only  the  overlapping  parts  of  two
strings.

Problem

An  overlap  alignment  between  two  strings    and    is  a  local  alignment  of  a  suffix  of    with  a  prefix  of
.  An  optimal  overlap  alignment  will  therefore  maximize  an  alignment  score  over  all  such  substrings  of  
 and   .

The  term  "overlap  alignment"  has  also  been  used  to  describe  what  Rosalind  defines  as  a  semiglobal
alignment.  See  “Semiglobal  Alignment”  for  details.

Given:  Two  DNA  strings    and    in  FASTA  format,  each  having  length  at  most  10  kbp.
Return:  The  score  of  an  optimal  overlap  alignment  of    and   ,  followed  by  an  alignment  of  a  suffix  
 of    and  a  prefix    of    achieving  this  optimal  score.  Use  an  alignment  score  in  which  matching
symbols  count  +1,  substitutions  count  -­2,  and  there  is  a  linear  gap  penalty  of  2.  If  multiple  optimal
alignments  exist,  then  you  may  return  any  one.
 Sample  Dataset

>Rosalind_54
CTAAGGGATTCCGGTAATTAGACAG
>Rosalind_45
ATAGACCATATGTCAGTGACTGTGTAA

Sample  Output

1
ATTAGAC-­‐AG
AT-­‐AGACCAT

Citation

This  problem  follows  Jones  &  Pevzner,  An  Introduction  to  Bioinformatics  Algorithms,  problem  6.22.

Problem  104
Quartet  Distance

Another  Tree  Distance

In  “Phylogeny  Comparison  with  Split  Distance”,  we  examined  the  split  distance  for
comparison  of  different  phylogenies  on  the  same  collection  of  taxa.

Yet  quartet-­based  phylogeny  offers  another  way  in  which  two  phylogenies  can  be
compared  (see  “Quartets”  and  “Counting  Quartets”).  Specifically,  we  wonder  how  many  quartets
can  be  inferred  from  one  tree  but  not  inferred  from  the  other.

Problem

In  “Counting  Quartets”,  we  found  an  expression  for   ,  the  number  of  quartets  that  can  be  inferred
from  an  unrooted  binary  tree  containing    taxa.

If    and    are  both  unrooted  binary  trees  on  the  same    taxa,  then  we  now  let    denote
the  number  of  inferred  quartets  that  are  common  to  both  trees.  The  quartet  distance  between    and
,    is  the  number  of  quartets  that  are  only  inferred  from  one  of  the  trees.  More  precisely,  
.

Given:  A  list  containing    taxa  ( )  and  two  unrooted  binary  trees    and    on  the  given
taxa.  Both    and    are  given  in  Newick  format.

Return:  The  quartet  distance   .

 Sample  Dataset

A  B  C  D  E
(A,C,((B,D),E));
(C,(B,D),(A,E));

Sample  Output

Problem  105
Finding  a  Motif  with  Modifications

Finding  Mutated  Motifs

We  have  discussed  at  length  the  importance  of  motif  finding  in  biology  for  genetic
strings.  However,  searching  for  exact  substring  matches  is  of  little  use  in
applications  because  a  motif  can  vary  under  the  effect  of  mutation.  Fortunately,
we  already  possess  functions  like  edit  distance  for  quantifying  the  similarity  of  two
strings.

Furthermore,  recall  that  each  chromosome  is  made  up  of  a  large  number  of  genes  (on  average,
each  human  chromosome  contains  over  1,000  genes).  Therefore,  to  determine  whether  a  newly
sequenced  chromosome  contains  a  given  gene,  neither  local  nor  global  alignment  applies.

One  possible  alignment  variant  for  finding  genes  is  semiglobal  alignment,  which  we  discuss  in
“Semiglobal  Alignment”;;  yet  semiglobal  alignment  only  allows  us  to  disregard  gaps  at  the  end  of
the  alignment.  To  find  a  known  gene  in  a  new  chromosome,  we  need  to  instead  align  the  gene
against  intervals  of  the  chromosome,  a  problem  that  calls  for  an  entirely  new  algorithmic  variation  of
alignment.

Problem

Given  a  string    and  a  motif   ,  an  alignment  of  a  substring  of  

against  all  of    is  called  a  fitting  alignment.  Our  aim  is  to
find  a  substring    of    that  maximizes  an  alignment  score
with  respect  to   . Figure  1.  Global,  local,  and  fitting
alignments  of  strings  v  =
Note  that  more  than  one  such  substring  of    may  exist, GTAGGCTTAAGGTTA  and  w  =
depending  on  the  particular  strings  and  alignment  score  used. TAGATA  with  respect  to  mismatch
One  candidate  for  scoring  function  is  the  one  derived  from  edit score.  Note  that  in  the  fitting
alignment,  a  substring  of  v  must  be
distance;;  In  this  problem,  we  will  consider  a  slightly  different aligned  against  all  of  w.  Taken  from
alignment  score,  in  which  all  matched  symbols  count  as  +1 Jones  &  Pevzner,  An  Introduction  to
Bioinformatics  Algorithms
and  all  mismatched  symbols  (including  insertions  and
deletions)  receive  a  cost  of  -­1.  Let's  call  this  scoring  function
 the  mismatch  score.  See  Figure  1  for  a  comparison  of  global,  local,  and  fitting  alignments  with
respect  to  mismatch  score.

Given:  Two  DNA  strings    and   ,  where    has  length  at  most  10  kbp  and    represents  a  motif  of
length  at  most  1  kbp.

Return:  An  optimal  fitting  alignment  score  with  respect  to  the  mismatch  score  defined  above,
followed  by  an  optimal  fitting  alignment  of  a  substring  of    against   .  If  multiple  such  alignments  exist,
then  you  may  output  any  one.

Sample  Dataset

>Rosalind_54
GCAAACCATAAGCCCTACGTGCCGCCTGTTTAAACTCGCGAACTGAATCTTCTGCTTCACGGTGAAAGTAC
CACAATGGTATCACACCCCAAGGAAAC
>Rosalind_46
GCCGTCAGGCTGGTGTCCG

Sample  Output

5
ACCATAAGCCCTACGTG-­‐CCG
GCCGTCAGGC-­‐TG-­‐GTGTCCG

Citation

This  problem  follows  Jones  &  Pevzner,  An  Introduction  to  Bioinformatics  Algorithms,  Problem  6.23.

Problem  106
Semiglobal  Alignment

Gaps  on  the  Ends  are  Free

We  have  covered  both  global  and  local  alignments.  However,  sometimes  we  need
a  hybrid  approach  that  avoids  the  weaknesses  of  these  two  methods.  One  such
alternate  approach  is  that  of  fitting  alignments  outlined  in  “Finding  a  Motif  with
Modifications”.

Another  tactic  is  to  allow  ourselves  to  trim  off  gaps  appearing  on  the  ends  of  a  global  alignment  for
free;;  this  is  relevant  if  one  of  our  strings  to  be  aligned  happens  to  contain  additional  symbols  on  the
ends  that  are  not  relevant  for  the  particular  alignment  at  hand.

Problem
 A  semiglobal  alignment  of  strings    and    is  an  alignment  in  which  any  gaps  appearing  as  prefixes
or  suffixes  of    and    do  not  contribute  to  the  alignment  score.

Semiglobal  alignment  has  sometimes  also  been  called  "overlap  alignment".  Rosalind  defines  overlap
alignment  differently  (see  “Overlap  Alignment”).

Given:  Two  DNA  strings    and    in  FASTA  format,  each  having  length  at  most  10  kbp.
Return:  The  maximum  semiglobal  alignment  score  of    and   ,  followed  by  an  alignment  of    and  
achieving  this  maximum  score.  Use  an  alignment  score  in  which  matching  symbols  count  +1,
substitutions  count  -­1,  and  there  is  a  linear  gap  penalty  of  1.  If  multiple  optimal  alignments  exist,  then
you  may  return  any  one.

Sample  Dataset

>Rosalind_79
CAGCACTTGGATTCTCGG
>Rosalind_98
CAGCGTGG

Sample  Output

4
CAGCA-­‐CTTGGATTCTCGG
-­‐-­‐-­‐CAGCGTGG-­‐-­‐-­‐-­‐-­‐-­‐-­‐-­‐

Citation

This  problem  follows  Jones  &  Pevzner,  An  Introduction  to  Bioinformatics  Algorithms,  Problem  6.24.

Problem  107
Finding  All  Similar  Motifs

The  Case  of  Mutated  Repeats

In  “Finding  a  Motif  with  Modifications”,  we  considered  a  problem  in  which  we  were
given  a  motif  and  a  long  string  (perhaps  representing  a  genome),  and  we  aimed  to
find  the  "closest"  substring  of  the  long  string  to  the  motif.  In  that  problem,
"closest"  was  defined  as  a  minimum  with  respect  to  edit  distance.

Yet  there  may  be  multiple  substring  candidates  from  the  genome  that  achieve  the  minimum
distance  to  the  motif;;  this  situation  might  occur  in  practice  when  the  motif  forms  a  repeat  that
occurs  multiple  times  with  variations  deriving  from  mutations.

In  this  problem,  we  would  like  to  find  all  substrings  of  a  genome  that  are  within  a  certain  fixed
distance  of  the  desired  motif.
 Problem

Given:  A  positive  integer    ( ),  a  DNA  string    of  length  at  most  5  kbp  representing  a  motif,
and  a  DNA  string    of  length  at  most  50  kbp  representing  a  genome.

Return:  All  substrings    of    such  that  the  edit  distance    is  less  than  or  equal  to   .  Each
substring  should  be  encoded  by  a  pair  containing  its  location  in    followed  by  its  length.

Sample  Dataset

2
ACGTAG
ACGGATCGGCATCGT

Sample  Output

1  4
1  5
1  6

Problem  108
Local  Alignment  with  Affine  Gap  Penalty

Building  Upon  Local  Alignments

We  have  thus  far  worked  with  local  alignments  with  a  linear  gap  penalty  and
global  alignments  with  affine  gap  penalties  (see  “Local  Alignment  with  Scoring
Matrix”  and  “Global  Alignment  with  Scoring  Matrix  and  Affine  Gap  Penalty”).

It  is  only  natural  to  take  the  intersection  of  these  two  problems  and  find  an  optimal  local  alignment
given  an  affine  gap  penalty.

Problem

Given:  Two  protein  strings    and    in  FASTA  format  (each  having  length  at  most  10,000  aa).
Return:  The  maximum  local  alignment  score  of    and   ,  followed  by  substrings    and    of    and   ,
respectively,  that  correspond  to  the  optimal  local  alignment  of    and   .  Use:

The  BLOSUM62  scoring  matrix.

Gap  opening  penalty  equal  to  11.
Gap  extension  penalty  equal  to  1.

If  multiple  solutions  exist,  then  you  may  output  any  one.

 Sample  Dataset

>Rosalind_8
PLEASANTLY
>Rosalind_18
MEANLY

Sample  Output

12
LEAS
MEAN

Problem  109
Isolating  Symbols  in  Alignments

How  Much  Does  it  Cost  to  Align  Two  Symbols?

As  we  saw  in  “Counting  Optimal  Alignments”,  there  will  usually  be  a  huge  number
of  different  optimal  alignments  of  two  given  strings.  In  this  problem,  which
represents  a  first  attempt  to  understand  how  much  optimal  alignments  can  differ,
we  will  select  two  symbols  at  a  time  from  the  two  strings  and  ask  how  much  the
maximum  alignment  score  can  differ  from  the  optimal  score  if  we  demand  that  these  two  symbols
must  be  aligned  (i.e.,  implying  that  one  symbol  must  be  substituted  for  the  other).

Problem

Say  that  we  have  two  strings    and    of  respective  lengths    and    and  an  alignment  score.  Let's
define  a  matrix    corresponding  to    and    by  setting    equal  to  the  maximum  score  of  any
alignment  that  aligns    with   .  So  each  entry  in    can  be  equal  to  at  most  the  maximum  score
of  any  alignment  of    and   .

Given:  Two  DNA  strings    and    in  FASTA  format,  each  having  length  at  most  1000  bp.
Return:  The  maximum  alignment  score  of  a  global  alignment  of    and   ,  followed  by  the  sum  of  all
elements  of  the  matrix    corresponding  to    and    that  was  defined  above.  Apply  the  mismatch  score
introduced  in  “Finding  a  Motif  with  Modifications”.

Sample  Dataset

>Rosalind_35
ATAGATA
>Rosalind_5
 ACAGGTA

Sample  Output

3
-­‐139

Citation

This  problem  follows  Jones  &  Pevzner,  An  Introduction  to  Bioinformatics  Algorithms,  Problem  6.21

Hint

For  the  sample  dataset  

Problem  110
Identifying  Reversing  Substitutions

Reversions  Complicate  Phylogenies

In  “Fixing  an  Inconsistent  Character  Set”,

we  mentioned  how  the  construction  of  a
phylogeny  can  be  complicated  by  a  the
reversion  of  a  character  to  a  past  state.
For  that  matter,  in  calculating  Hamming  distance  and  edit
distance,  the  assumption  of  parsimony  required  us  to
assume  that  if  some  nucleotide  base  or  amino  acid  is
aligned  with  an  identical  symbol  in  two  genetic  strings,
then  it  has  not  changed  on  the  evolutionary  path  between
the  two  taxa.

However,  this  model  is  too  strict  in  practice,  where  a  base Figure  1.  Illustration  of  an  amino
or  amino  acid  can  change  to  another  state  and  then acid's  reversing  substitution  after
two  point  mutations.
change  back  as  the  result  of  two  point  mutations,  which  is
called  a  reversing  substitution;;  see  Figure  1.  In  the  case
of  DNA,  the  presence  of  only  four  bases  makes  randomly  occurring  reversing  substitutions
common;;  these  substitutions  will  carry  over  into  amino  acid  language  via  transcription  and
 translation.

Unfortunately,  with  the  possible  exception  of  experimental  evolution  in  bacteria,  we  lack  the  luxury
of  knowing  the  ancestral  state  of  a  nucleic  acid  strand.  Instead,  we  must  infer  its  most  probable
ancestor  from  homologous  strands.  To  do  so,  we  may  use  characters  to  construct  a  phylogeny
(see  “Character-­Based  Phylogeny”),  then  apply  alignment-­based  phylogeny  to  infer  strings  for  the
tree's  internal  nodes  (see  “Alignment-­Based  Phylogeny”).  Only  once  we  have  an  adequate  picture
of  the  entire  phylogeny,  including  its  internal  nodes,  can  we  hope  to  identify  reversing  substitutions.

Problem

For  a  rooted  tree    whose  internal  nodes  are  labeled  with  genetic  strings,  our  goal  is  to  identify
reversing  substitutions  in   .  Assuming  that  all  the  strings  of    have  the  same  length,  a  reversing
substitution  is  defined  formally  as  two  parent-­child  string  pairs    and    along  with  a  position
index   ,  where:

there  is  a  path  in    from    down  to   ;;

;;  and
if    is  on  the  path  connecting    to   ,  then   .

In  other  words,  the  third  condition  demands  that  a  reversing  substitution  must  be  contiguous:  no  other
substitutions  can  appear  between  the  initial  and  reversing  substitution.

Given:  A  rooted  binary  tree    with  labeled  nodes  in  Newick  format,  followed  by  a  collection  of  at
most  100  DNA  strings  in  FASTA  format  whose  labels  correspond  to  the  labels  of   .  We  will  assume
that  the  DNA  strings  have  the  same  length,  which  does  not  exceed  400  bp).

Return:  A  list  of  all  reversing  substitutions  in    (in  any  order),  with  each  substitution  encoded  by
the  following  three  items:

the  name  of  the  species  in  which  the  symbol  is  first  changed,  followed  by  the  name  of  the  species
in  which  it  changes  back  to  its  original  state
the  position  in  the  string  at  which  the  reversing  substitution  occurs;;  and
the  reversing  substitution  in  the  form  original_symbol-­>substituted_symbol-­>reverted_symbol.

Sample  Dataset

(((ostrich,cat)rat,mouse)dog,elephant)robot;
>robot
AATTG
>dog
GGGCA
>mouse
AAGAC
>rat
GTTGT
>cat
GAGGC
>ostrich
GTGTC
>elephant
AATTC

Sample  Output
dog  mouse  1  A-­‐>G-­‐>A
dog  mouse  2  A-­‐>G-­‐>A
rat  ostrich  3  G-­‐>T-­‐>G
rat  cat  3  G-­‐>T-­‐>G
dog  rat  3  T-­‐>G-­‐>T