Phylogenetic Inference

1
Phylogenetic Inference

• Methods

2
need optimality criteria algorithm to
search for the best tree given the optimality
criteria
3
Best tree OR True tree
4
Types of optimality criteria used to infer
phylogeny from sequence

• Distance methods
• Parsimony
• Likelihood
• Probabilistic methods
• Phylogentic invariants

5
Distance based methods

• Minimum Evolution Principal
• The tree with the smallest sum of branch lengths
  is the best tree

6
A
B
r
s
t
u
v
D
C
dAB dCD dAD dBC r s u v r t v
s t u 2r 2s 2u 2v 2t
T r s u v t T (dAB
dCD dAD dBC ) / 2
7
Number of possible unrooted trees from n
sequences
e.g. for 20 sequences there are approximately 1020
8
For realistic numbers of sequences it is
impossible to consider all possible trees. Need
algorithms that can arrive at the best tree
without considering all possible trees.
9
Neighbour joining is an approximation to minimum
evolution
10
Neighbour Joining
8
7
1
6
2
3
5
4
Choose the pair that minimizes the length of the
resulting tree
11
Distance methods

• Calculate the distance CORRECTING FOR MULTIPLE
  HITS
• The Distance Matrix
• 7
• Rat 0.0000 0.0646 0.1434 0.1456
  0.3213 0.3213 0.7018
• Mouse 0.0646 0.0000 0.1716 0.1743
  0.3253 0.3743 0.7673
• Rabbit 0.1434 0.1716 0.0000 0.0649
  0.3582 0.3385 0.7522
• Human 0.1456 0.1743 0.0649 0.0000
  0.3299 0.2915 0.7116
• Oppossum 0.3213 0.3253 0.3582 0.3299
  0.0000 0.3279 0.6653
• Chicken 0.3213 0.3743 0.3385 0.2915
  0.3279 0.0000 0.5721
• Frog 0.7018 0.7673 0.7522 0.7116
  0.6653 0.5721 0.0000

12
Correction for multiple hits

• A great many models used for nucleotide sequences
  (e.g. JC, K2P, HKY, Rev, Maximum Likelihood)
• aa sequences are even more complicated!
• Can take account of different rates of evolution
  at sites (e.g. gamma distribution)
• Accuracy falls off drastically for highly
  divergent sequences
• Is it necessary to use the most realistic model??

13
The most accurate nucleotide substitution model
doesnt necessarily give the best estimate of the
true tree - models with higher numbers of
parameters provide distance estimates with higher
variance
14
How to infer the true tree? How to keep
reviewers/editors happy?
15
In short distance methods

• Can be fast and simple
• e.g. UPGMA, Neighbour Joining, Minimum Evolution,
  Fitch-Margoliash

16
Maximum Parsimony

• Occams Razor
• Entia non sunt multiplicanda praeter
  necessitatem.
• William of Occam (1300-1349)

The best tree is the one which requires the least
number of substitutions
17

• Check each topology
• Count the minimum number of changes required to
  explain the data
• Choose the tree with the smallest number of
  changes
• Usually performs well with closely related
  sequences but often performs badly with very
  distantly related sequences
• With distantly related sequences homoplasy
  becomes a major problem

18
Not all trees need to be considered (branch and
bound method still guarantees to find MP
tree) In practice a heuristic search is often
performed (involving branch swapping e.g. NNI,
SPR, TBR). No guarantee of finding the MP tree.
19
Long Branches Attract

• In a set of sequences evolving at different
  rates the sequences evolving rapidly are drawn
  together

20
Comparison of methods

• Inconsistency
• Neighbour Joining (NJ) is very fast but depends
  on accurate estimates of distance. This is more
  difficult with very divergent data
• NJ can suffer from Long Branch Attraction
• Parsimony suffers from Long Branch Attraction.
  This may be a particular problem for very
  divergent data
• Parsimony can be computationally intensive
• Codon usage bias can be a problem for MP and NJ
• NJ and MP both perform well if sequences are not
  too divergent
• Maximum Likelihood can the most reliable but
  depends on the choice of model and can be very
  slow
• Methods may be combined

21
(No Transcript)
22
The Molecular Clock

• For a given protein the rate of sequence
  evolution is approximately constant across
  lineages
• Zuckerkandl and Pauling (1965)

This would allow speciation and duplication
events to be dated accurately based on molecular
data
Local and approximate molecular clocks more
reasonable
23
Relative Rate Test

• Test whether sets of sequences are evolving at
  equal rates (local molecular clock hypothesis)

e.g. RRTree, Robinson-Rechavi http//pbil.univ-lyo
n1.fr/software/rrtree.html