BT 3040 Endrem

Part II

④
n

Mol weight = E mcri )

f- 1

when i c- C ) len
is
of sequenced
' n
,
,n
Ji is the ith residue
mcri molwt ith residue
) is the
of
i.
Molest =
146 + 147 + 146 t 181 + 174 + 181
K E 9 Y R Y

Molwt = 975 arm 2 97T dalton

n I
-

② Base stacks
energy
=

i.
Ee critic , )
1
/(n -

1)

when i c- C ) len
is
of sequenced
' n
,
,n
Ji is the ith residue

e. Citi 8in ) is the bare two

energy of
consecutive residues d-
position 's

(GTA)/ 8 + -9

AG
+ -9

GA
+ -9

AG
+ -10

GC

thy =
-
9

③
①M
I

t
L

A
-
I

t
T

S
①
LA

-
D
I

-
T
-
W

L
+
k

R A
+
V

(From Blosum )
62

Identity = 2110 = 2070

similarity = 6110 =
60%
&

I •

] • •

1- .

> • •

I •

o
o

→ • •

I •

H
① •

2 0

V H L E E K V T A L W G K

HLEE by VT are aligned

Part 11T
=

QI
⑨ A database is an organized collection
of information in a

computer
-

compatible format
⑤ Factors
Content
:

Fin the
-
:
content according to target audience .
If
but database
target audience
biologist
are
will
has
it
information
belated then
astronomy
to one me
no
.

the other
Knowing content also helps create the
databases as the schema
feouetuus of
the such ,
the
etc
user
interface ,
.

Ontology :
Ontology the
description of terms cesed Many
is .

a twin for systematic storage of data lot of

,
a
,

technical terms to avoid

cued confusion Also
are .

sin the database is to be accessible to people of all

 levels of knowledge ,
even new students . Hence to make
is
the database
user-friendly and usable ,
ontology my
linportant .

Kia schema is the interrelationship within the data

: .

For eg a
proteus has
thermodynamic
:

sequence ,

property ,
chemical
properties function etc , ,
.

how true various teens are connected forms the

and hence it crucial role

schema
plays a .

torment To avoid defined template / for met

confusion
:
a
,

should be used to prevent the data otherwise ,

user will get confined when accosting multiple some .

retrieval :
Fuially ,
tu ever must be able to retrieve the

G pinnae - data
according to his / her need . Hence an

interface must be
provide and some DBMS
like
query language Sql must be used

Fwtherueauh If interested in kenowiy more than

is
:
a urn

what's
given in our database we should
provide ,

bike to onn database with different information

Biology as a subject
very complicated ranging from is
,
the
there
many relationships to factor so many
micro to macro are so , ,

unanswered question and so much data to titer through

there is with
Hence a lot
of research being done in the ticket ,

each
group conducting their own enperimevto , creating their own

algorithms net .

This database for

is
why we see so
many
various
categories .

They contain
mostly the same information ,
but also somewhat

slightly different .

This and
very competitive forward
research
makes there
groups pushes .

Bio databases contain a lot of information and hence structured

and retrieval
storage are
key .

Given for a single entity , g.

a
protein ,
there is so much divers
information , eg .
thermal
properties chemical properties having
,
,
a

relational database
,
i. e
,
multiple datahees containing specific information
such as the various
properties ,
and the relations bring them
together .
?⃝
⑨ European Molecular
Biology lab

DNA Data Bank of Japan

②
⑨ A :
GTGCTGCACG di ,j =
Number
of mismatches between
B :
A G C 1- GC A AEC ith { jtn seq
C :
GT G C T G C ACT
☐ : C T C T G C A G A A

E : GT A T C A C AT A

A B C D
B 9

c
① 9

9 5
A. :
☐ 9
B D
C- 6 8 6 7 E °
°

(A. c) O
o
0

CB D)
(
,
B
AC D
(A. D. E) 0

B 9 0

① A. D. E) @ D)
⑧
. -

D 9

E 6 8 7

AC BD

BD 9

€ ⑥ 75

(✗ A. c) E)
, ,(B. D ))
④ I :
GTGCTGCACG
II :
AGC 1- GC A ACC
It :
GTGCTGC ACT
TI : CTC TG CA GAA
I : GT A TCA C AT A

1 2 3 4 5 6 7 8 9 10

A 1 0 I 0 0 I 2 4 I 2
G 3 I 2 0 2 2 •
I 1 I
C 2 3
I
?
I 0 2 2 0
I
T O 4 0 3 2 O O O ,

'
DNA
,
hence ,
p= 14,111--5

win
. =
in in base ,jtn position
N : number
of sequin

I 2 3 4 5 6 7 8 9 10

A -0.18 -1.79 -0.18 -1.79 -1-79 -0.18 0-41 1.0A -0.18 0.41
G o . >> -0.18 0.41 -1.79 0.41 0.91 -1-79 -0.18 -0.18 -0.18
C 0.18 1.79 0.41 -10.18 0.41 0.77 -1.79 041 0.18
- -

0.41 -

T -
I -79 1-04-1.79 0.77 0.41 -1-79 -1.79 -1.79 -0.18 -0.18

② Pos 3 Posh

I G C

I C G

IHI G C

II C A

¥ A T

3 9
A 0-2 012
G 0.4 0.2
C 0.4 0.4
T O 80.2
 member of see
Éf J
:
n

Entropy
: ; lnf .

in residue / bare
i. 1
frequency ogih entity
at
fi
-
:

given position
posts
=
0-2 / no -2 1- 0.4 Ch 0.9 1- 0.4 In 0.4

= -1.055

to
= 0 -

21h0 -2 1- Or 41h0 -4 t 0-2140.2 1- 0 .

relief 2

= -1.33

⑨ -
-

O
G
O
T
O
G
O
C

O
T

O
G
O
C
O
A
O
CT

O O

C O O O O 2 0 0 2 0 2 ☐

4 I 0 A
T
⑨
O O 2 • 0 &

C O O O l ② A 3 2 0 3 0

T O O 2 o o ⑧ O O O O 5
G O 2 0 4 1 I ⑥ 3 0 0 2
C O O r r 6 3 3 ⑧ 5 2 •

A O O O O 3 5 2
55 ④ > *

G O 2 8 2 00 2 7 4 7 ⑨ 6
A 0 I 6 6 80
0 0
I • 4 6
A O O O O O O
1 3 3 3 ⑤

Gt GC TG CA C - T
CTG CA GA A

⑧ When it to data driven modeling using AI methods

⑨ comes
,

the data
plays any significant role

whether it be black boar like net

a
deep neural or a

statistics driven bayesian net ,

the data can determine
whether model screeds or not .

In this
regard ,
the
preprocessing of data
plays a huge
role .

Prepnocesiy here can mean

normalization ,
filthy ,
 augmentation ,
etc .
Normalization is
very uniportant ,

otherwise the Tae

input value can skew tire model .

For the model trainee value C- Co D

eg if is or
-

and suddenly three is an

input > 1
,
that will confine tree model

find spurious correlations

and
may push it to .

Filtering could mean

removing repeated samples if samples ,
are

it bias the
repeated can model .

Augmentation means
slightly altering the input .
Sine neural welt
to
have a
tendency orufit to training data distribution
, augmentative
confuse the model a little bit and umehe it more robust .

Eg adding :
some noise to input feature vector .

neural
Sample size , deep aece
highly dependent on

data
of samples are small
the
sample size .

If number
then the model can
easily ovefit Smaller data sige
gently
but this leads
.

models with smaller parameters

require ,

to weather generalisation .

Dumeirsionaliiz reduction models form

.
In which
hyperplanes
for clarification such as SVM s
,
having redundant feather dimension
the model Using methods such as PCA t.SN e.
can
confine .
,

the
dimension be
Auto encodes
,
can reduced to keep only most valuable

Ones ,

is
After date
prepared we can use
multiple methods

For regression ,
we can ve

linear models : Find linen relationship

between
the
and
reconstruction
data
prediction
lose .
by minimize

☐ NN :
deep newel nets are
very versatile
but are
contingent
data available in bio, the
on amount
of . Sin ,
ground
and
Casmnhoiud
truth available is
really spans hence we cannot
network and
above ) have
very deep hence lose out on

perfoormau Unsupervised learning Wwe

ground truth is not
-

required key but is a very difficult tale and lot

is .
a

of nevaeh
is
being put into find methods of unsupervised leering .

Since DNN are black bone by nature it is very difficult

to intuit it's working .

Classification .
 boned
True based method :
bayesian
methods whee tree
are constructed between

input and ernpeeted class . The tires are

modelled
and
on the basis
of probability
Lena have
good enplain ability behind their
actions

SVM :

finds hyperplanes between Various classes

by
minimising the distance of the
predicted plan from
class
each point
of a .

DNN :
DNNS ear
again be wed men .

Clustering based methods line K -

mean
,
DBS CAN ,
etc .

be used well classify

can as to on the basis
of the
clusters made -

Verification of model

K -

told chose validation is the staler method

of
testing the
capacity of the model . The data is split
into train & tent
splits . The model is trained on train split
and evaluated on test .

test split distribution

In
genal it is a
good practice to have
to be
slightly different from train
split .

typeepaiauuhe Gone details

hypenpaa meter optimization is a

huge
tart in
too gsfindis
DNA ,

the rate and

right leering schedule , weigh initihgidsitoir Alongside .

also huge
the
type of optimizes such as SGD or Adam
plays a

role

④ N
= >5
,
AP = 12

=
p 13

[ correctly
Tp = >
predicted position)
☒ p= 6 [income thy predictive positive]
[ incorrectly
predicted negative ]
F N = 6

TN = 56 [ correctly predicted negative]

 TP
sensitivity =

¥-74

¥ 0.583
= =

specificity ,¥p
=

§÷
= = 0 -903

accuracy TP+m
I

= 63 = 0.84
-75

3 9- 25

i.u.in/.....iM/
⑨ 5 o a • • ~

, , .

2 to

hydrophobicity profile ,
plot of hydrophobicity values
us residue
position

Sey : A I KSWVKTIARTYLLNS
grain

poweo Of Anrphiphetiits = E
/( ait , , 9in -

@ its , air /
N

= 2 - O l