I519 Introduction to Bioinformatics, 2012

Phylogenetic Tree Reconstruction

 Evolution theory
 Speciation
– Evolution of new organisms is driven by
• Mutations
– The DNA sequence can be changed due to single base changes, deletion/insertion
of DNA segments, etc.
• Selection bias
– Speciation events lead to creation of different species.
– Speciation caused by physical separation into groups where
different genetic variants become dominant
 Any two species share a (possibly distant) common
ancestor
 The molecular clock hypothesis
 Indiana University (Michael Lynch, Jeff Palmer, Matt
Hann et al)
Lynch: The Origins of Genome Complexity
 Phylogeny (phylogenetic tree)
 A phylogenetic tree is a graph reflecting the approximate
distances between a set of objects (species, genes,
proteins, families) in a hierarchical fashion

taxon

Aardvark Bison Chimp Dog Elephant

 Leaves – current species; sequences in current species
 Internal nodes - hypothetical common ancestors
 Branches (Edges) length - “time” from one speciation to the next
(branching represents speciation into two new species)
 Phylogenetic trees (binary trees)
Rooted tree Rooted tree satisfying “molecular clock”
hypothesis: all leaves at same distance from the root.
root
6 root
8 7 time
7
6
8
3
1 2 5
4 1 2 3 4 5 UMPGA

Unrooted tree: Note: 1-5 are called leaves, or

3
leaf nodes. 6-8 are inferred nodes
4
corresponding to ancestral species or
8
6
molecules. Branches are also called
2 7 5 edges. The edge lengths reflect
evolutionary distances.
1
 Morphological vs. molecular
 Classical phylogenetic analysis: morphological
features: presence or absence of fins, number of
legs, lengths of legs, etc.

 Modern biological methods allow to use

molecular features
– Gene sequences
– Protein sequences

 Species tree and gene tree

From sequences to phylogenetic tree
Rat QEPGGLVVPPTDA..
Rabbit QEPGGMVVPPTDA..
Gorilla QEPGGLVVPPTDA..
Cat REPGGLVVPPTEG..

There are many possible types of

sequences to use (e.g. Mitochondrial vs
Nuclear proteins).
 How to choose the best tree?
 To decide which tree is best we can use an optimality
criterion.
 Parsimony is one such criterion (the other criteria:
Maximum likelihood, minimum evolution, bayesian)
 It chooses the tree which requires the fewest mutations
to explain the data.
 The Principle of Parsimony is the general scientific
principle that accepts the simplest of two explanations
as preferable.
 Principle of Parsimony
Parsimony based methods look for a tree with the
minimum total number of substitutions of symbols
between species and their ancestor in the phylogenetic
tree.
AAA 1 AAA
AAA AGA AAA AAA
1 1 1 1 2
AAA AGA AGA GGA
AAG GGA AAG AAA
Total #substitutions = 3 Total #substitutions = 4

The left tree is preferred over the right tree.

 Maximum parsimony
S1 CACCCCTT
S2 AACCCCAT
S3 CACTGCTT
S4 AACTGCTA
(S1,S2),(S3,S4) 20011011 6 
(S1,S3),(S2,S4) 10022011 7
S1 C C S3 S1 C A S2

C A

S2 A mutation=2 mutation=1
A S4 S3 C A S4
(S1,S2), (S3, S4) (S1,S3), (S2, S4)
We can EASILY get different trees
(the “reality check” paper)

 Input sequences
 Multiple alignment programs
 Substitution models
 Phylogenetic tree reconstruction methods
Trees – what might they mean?
Species A
Species B
Species tree
Species C

Species D

Seq A
Gene tree
Seq D
Seq C
Seq B
 Lack of resolution

Seq A
Seq D
Seq C

Seq B

Weak support, 40% bootstrap for bipartition (AD)(CB)

(typical >80%)
Long branch attraction (LBA)
the two longest branches join together
Seq A
Seq D
Seq C

Seq B

Strong support, e.g., 100% bootstrap for (AD)(CB)

 Horizontal gene transfer
species A
species B
Species tree Gene Transfer
species C
species D

Seq A
Gene tree
Seq D
Seq C

speciation Seq B
gene transfer
 Gene duplication & loss
species A
Species tree
species B
species C
Duplication species D
Seq A
Loss
 Seq B
Gene tree  Seq C A
D
Seq D C
Seq B’
B
Seq C’
 Seq D’
 Orthologs and paralogs
(important for function annotation)
Seq A
Orthologs:
Seq B sequences diverged after
a speciation event
Seq C
Paralogs:
Seq D sequences diverged after
gene Seq B a duplication event
duplication Xenologs:
Seq C
sequences diverged after
Seq D a horizontal transfer

Duplicated genes may have different functions!!

 Phylogeny (phylogenetic tree)
reconstruction: overview
 Tree topology & branch lengths
 Computational challenge
– Huge number of tree topology
3 sequences: 1 (unrooted)
4 sequences: 3
5 sequences: 15
10 sequences: 2,027,025
20 sequences: 221,643,095,476,699,771,875
n sequences (unrooted & rooted) ??
 Most methods are heuristic
 Two types of methods
– Distance based (input: distance matrix; UPGMA & NJ)
– Character based (input: multiple alignment)
Models of evolutionary distance
1. Simplest case: Jukes-Cantor model
-- equal probability of change to any nucleotide

2. Other models take into account transitions vs. transversion frequencies

-- Kimura: different probabilities for transitions, transversions
-- HKY: different probabilities for transitions, transversions &
takes into account genomic nucleotide biases
Transition: R to R
b Y to Y
A G

  Transversion: R to Y
Y to R
C T
b where R = A,G
Y = C,T
 Distance based phylogeny
reconstruction
 Phylogeny reconstruction for 3 sequences is
EASY
– There is a single tree topology
– The branch lengths can be calculated as follows:
To compute: branch lengths a, b and c, such that

A
a
c C
Input: DAB, DBC and DAC (pairwise distances) b
B
Output:
 Fitch-Margoliash (FM) algorithm
 For phylogeny reconstruction with more than 3 sequences
 For example, given 5 sequences, A, B, C, D and E. The
tree can be reconstructed as follows
– First choose the closest sequence pair, suppose it is D and E
(based on the input pairwise distances; e.g., DDE=10)
– To calculate the branch lengths from D and E to their common
ancestor (denoted as d and e), we combine the remaining
three sequences (A, B and C) and treat them as a single
composite sequence (and define and so on)
-- so again we are dealing with 3 sequences, and we can
easily calculate the branch lengths
– Then merge D and E into a cluster and treat it as a composite
sequence, and update the distance table so that
and so on.
– Repeat the above steps until no more clusters to merge
 Distance based method: UPGMA
 UPGMA: Unweighted Pair Group Method with Arithmetic
Mean
 Assume same rate evolution (molecular clock hypothesis)
 The length from root to each leaf is the same (ultra
metric).
 It is similar to Fitch-Margoliash algorithm (merge two most
similar sequences or clusters first); but the calculation of
branch lengths is even simpler.
 For example, for the same example shown above with
five sequences, d=e=5 (d and e are the branch lengths
from sequence D and E to their common ancestor)
 Neighbor-joining (NJ)
(Seitou & Nei algorithm)
 Minimum evolution -- the least total branch
length (distance-based)
 Bottom-up clustering method
 Does not assume same rate evolution
 Fast & produce reasonable trees
 NJ method
3 3

2 2
X 4 Y X 4

1 1
5 5

NJ looks for two sequences (clusters) to merge that minimizes:

Where

C1 and C2 are most similar to each other, while they are most
dissimilar to the other clusters (far from others)
 Comparison of FM, UPGMA and
NJ methods
 All are hierarchical clustering methods
 All define the distance between two clusters as the
average pairwise distance

 FM and NJ do not assume “molecular clock”; UPGMA

does and it uses a simpler way to calculate branch
lengths.
 UPGMA and FM choose and merge the closest sequence
(cluster) pair first, but NJ looks for two sequences
(clusters) that are not only close to each other (as in
UPGMA and FM) but also far apart from the rest
 Parsimony based reconstruction

1. A procedure to find the minimum number of

changes needed to explain the data for a given
tree topology, where species are assigned to
leaves. (Small Parsimony Problem)

2. A search through the space of trees. (hard

problem!) (Large Parsimony Problem)
 Small Parsimony Problem
 Compute the minimum number of mutations on a
GIVEN tree
 Fitch algorithm
 Sankoff algorithm (subtree; DP)
 The Fitch Algorithm
 Pick an arbitrary root to work towards (for unrooted tree)
 Work from the tips of the tree towards the root. Label each node with the
intersection of the states of its child nodes.
 If the intersection is empty label the node with the union and add one to the
cost
TC +1 T
+1
C C

AGC +1 Cost=4 C
CT +1 GC +1 C C
CT= +1 +1 +1
C T G C A T C T G C A T
Calculate Fitch score Internal node labeling
 Sankoff Algorithm
 More general than the Fitch algorithm.
 Assumes we have a table of costs cab for all possible
changes between states a and b (A, T, C or G for DNA)
 For each node i in the tree we compute S(i,a) the minimal
cost given that node i is assigned state a.
 In particular we can compute the minimum value over a
for S(root,a) which will be the cost of the tree.
 Sankoff algorithm: DP
G C T
inf inf 0 inf inf inf inf 0
A G inf 0 inf inf
T
0 inf inf inf inf inf 0 inf inf inf inf 0
S(i,A) = 
S(i,G)=0 S(i,T) = 0

1 4 1 4 4 1 4 1
Initialization:
S(i,a) = 0 i labeled by a, or
S(i,a) = 
2 5 1 5
5 2 5 1
a{A,T,C,G}
cab A C G T Iteration:
A 0 2 1 2 S(i,a) = minx(S(j,x)+c(a,x))
C 2 0 2 1 + miny(S(k,y)+c(a,y))
G 1 2 0 2 Termination:
5 5 4 4
T 2 1 2 0 minaS(root,a)
 Large Parsimony Problem
 The small parsimony problem – to find the score
of a given tree - can be solved in linear time in
the size of the tree.
 The large parsimony problem is to find the tree
with minimum score.
 It is known to be NP-Hard.
 Tree search strategies
 Exact search
– possible for small n only
 Branch and Bound
– Use “cleaver” rules to avoid some branches of trees
– up to ~20 (25) taxa

 Local search - Heuristics

– Pick a good starting tree and use moves within a
“neighbourhood” to find a better tree; e.g., nearest-neighbor
interchanges (NNIs)
 Meta-heuristics
– Genetic algorithms
– Simulated annealing
 Branch and bound

http://evolution.gs.washington.edu/gs541/2005/lecture25.pdf
Branch and bound

This branch
can be safely
“neglected”!
Nearest neighbor interchange
 Probabilistic approaches to
phylogeny
 Rank trees according to their likelihood
P(data|tree), or, posterior probability P(tree|data)
(Bayesian)
 Maximum likelihood methods
 Sampling methods
 Calculate likelihood
Felsenstein’s algorithm for likelihood
 Initialization: i=2n-1
 Recursion: Compute P(Li|a) for all a as follows
if i is leaf
P(Li|a)=1 if a is the label of the leaf, otherwise 0
else
Compute P(Lj|a), P(Lk|a) for all a i
P(Li|a)=b,cP(b|a,tj)P(Lj|b)P(c|a,tk)P(Lk|c) a
j k
b c
 How can one tell if a tree is
significant
Biological knowledge

Bootstrapping: how dependent is the tree on the dataset

1. Randomly choose n objects from your dataset of n, with replacement
(picking columns from the alignment at random with replacement)
2. Rebuild the tree based on the subset of the data
3. Repeat many (1,000 – 10,000) times
4. How often are the same children joined?
Jackknifing: how dependent is the tree on the dataset
1. Randomly choose k objects from your dataset of n, without replacement
2. Rebuild the tree based on the subset of the data
3. Repeat many (1,000 – 10,000) times
4. How often are the same children joined?
 Commonly used phylogeny
packages
 369 phylogeny packages
(http://evolution.gs.washington.edu/phylip/software.html) and 54
free servers (as of Sep 30, 2011)
– Phylip (general package, protdist, NJ, parsimony, maximum
likelihood, etc)
– PAUP (parsimony)
– PAML (maximum likelihood)
– TreePuzzle (quartet based)
– PhyML (maximum likelihood)
– MyBayes
– MEGA (biologist-centric)
 What a surprise
 “..because of their intrinsic reliance on summary
statistics, distance-matrix methods are assumed to
be less accurate than likelihood-based
approaches. In this paper [Science, 327:1376–
1379, 2010], pairwise sequence comparisons are
shown to be more powerful than previously
hypothesized. A statistical analysis of certain
distance-based techniques indicates that their data
requirement for large evolutionary trees essentially
matches the conjectured performance of maximum
likelihood methods—challenging the idea that
summary statistics lead to suboptimal analyses. ”
 Readings
 Chapter 7 (Revealing Evolutionary History)
 Chapter 8 (Building Phylogenetic Trees)