Phylogenetic Reconstruction Using NeighborJoining

1
Phylogenetic Reconstruction Using
Neighbor-Joining Maximum Likelihood Methods
With Different Evolutionary Models

• PROJECT PRESENTATION
• CAP5510, Fall 2004
• Rebecca Gray
• Dennis Klop
• Alexandra Martinez

2
Outline

• Problem Description
• Methodology
• Neighbor-Joining Algorithm
• Evolutionary Models
• Phylip Package for Maximum Likelihood
• Analysis
• Results
• Conclusions

3
Problem Description

• 24 sequences generated by Mulligan Lab.
• Sequences from mitochondrial d-loop region
• Average length of sequences 500 bp
• Exact evolutionary relationship between these
  sequences is unknown
• Our goal gt infer the correct phylogenetic
  relationship by using the most suitable model of
  evolution

4
Neighbor-Joining Method

• Reconstructs phylogenetic trees from evolutionary
  distance data.
• Provides both topology and branch lengths.
• Input multiple sequence alignment.
• Initial distance matrix is derived from msa.
• Starts with a starlike tree, and iterates through
  a series of clustering steps.
• At each clustering step, find the next pair of
  OTUs (neighbors) to join as the one that
  minimizes the sum of branch lengths.
• We implemented this algorithm in Java.

5
Evolutionary Models

• Jukes-Cantor model
• Simplest, equal base frequencies
• Free parameter rate of nucleotide substitution
• Jukes-Cantor corrected model
• By Mount, unequal base frequencies
• Free parameter rate of nucleotide substitution
• Kimura 2-Parameter model
• More complex, equal base frequencies
• Free parameters transition / transversion rates

6
The Phylip Package (1)

• Used for creating trees with the Maximum
  Likelihood method.
• DNAML ? DNAMLK
• DNAMLK assumes the molecular clock hypothesis
  under which proteins and nucleic acids evolve at
  constant rates through time and for different
  lineages
• User specified parameters
• Global rearrangements
• Using outgroups ? Trastrep sequences

7
The Phylip Package (2)

• SPR Subtree Pruning and Regrafting
• Identify and remove a subtree
• Reattach to each possible branch of the remaining
  tree
• Improves result since the position of every
  species is reconsidered
• High time complexity ? triples the runtime of the
  program!

8
The Phylip Package (3)

• Creating phylogenies of all the Trastrep
  sequences with or without the global
  rearrangements.
• Comparing Maximum Likelihood scores

9
The Phylip Package (4)

• Bootstrapping to obtain statistical support for
  our branches of our trees
• In DNAML JC, JC corrected K2-P model
• In DNAPARS creates a tree based on the maximum
  parsimony method
• Creating consensus trees out of a hundred
  alignments for each model
• Obtaining bootstrap values for each node for each
  model

10
Analysis of the Models of Evolution - Techniques

• 1. Evaluate trees based on biological intuition
• There are 6 different genuses represented
• We expect Trastrep to be placed as the outgroup
• Bubalus sequences should be placed together
• Bos sequences should be placed together
• 2. Select ML tree with highest likelihood score
• 3. Compare NJ trees with ML bootstrapped
  consensus trees
• The model of evolution that has the most
  similarities between the two methods is the best

11
Analysis - Creation of Bootstrapped Consensus
trees

• To determine which model of evolution best
  supported our data, we used the bootstrap method
• Non-parametric bootstrapping involves re-sampling
  the data a set number of times (in this case
  100x)
• The bootstrap analysis begins with the initial
  multiple alignment and re-creates this dataset
• Bootstrap values indicate the number of replicate
  data sets in which a particular branch was
  created
• There is not a set standard bootstrap value which
  indicates statistical support we chose the value
  of 70

12
Analysis - Consensus Tree

• Each bootstrapped data set, which now consists of
  100 different multiple alignments, was used as
  input into the PHYLIP programs DNAML and DNAPARS
• The parameters of DNAML were manipulated to
  reflect the particular model of evolution under
  review (JC, JC corrected, K2P) as discussed
  before DNAPARS does not require any parameter
  optimization
• For each of the four models, 100 trees were
  created
• The program CONSENS in the PHYLIP program was
  used to create a consensus tree for each of the
  100 trees for each model of evolution
• This consensus tree reflects the best topology
  that fits the 100 trees each branch is given
  bootstrap value

13
Analysis 1 Consensus Parsimony Tree

• Notice Trastrep not outgroup
• Notice Bubalus not grouped together

14
Analysis 1 Consensus Jukes-Cantor

• Notice that Bubalus is the outgroup
• Trastrep is nested within the tree

15
Analysis 1 ML Consensus Bootstrapped Kimura
2-Parameter

• Notice that Trastrep is nested in tree
• Bos sequences are separated

16
Analysis 1 ML Consensus Bootstrapped
Jukes-Cantor corrected

• Notice that Trastrep is placed as the outgroup
• All of the Bubalus species are placed together
• All of the Bos sequences are placed together,
  closer to Bubalus than Trastrep

17
Analysis 2 ML tree with highest likelihood score
(JC corrected)

• We created 3 ML trees in PHYLIP for each model of
  evolution
• The tree with the highest score was the JC
  corrected tree
• Jukes-Cantor
• 2787.4376
• Jukes-Cantor corrected
• 2740.70636
• Kimura 2-Parameter
• 2818.97662.

18
Analysis 3 Comparisons between NJ trees and ML
consensus bootstrapped trees

• 1. Calculated the number of shared branches
  between the NJ tree and the ML consensus tree for
  each model of evolution (Parsimony vs. Mismatch,
  Jukes-Cantor, Jukes-Cantor corrected, Kimura
  2-parameter
• 2. Calculated the number of shared branches
  between trees with gt70 bootstrap support
• 3. Calculated the number of shared internal
  branches between trees, i.e. those branches that
  do not just connect leaves
• 4. Calculated the number of shared internal
  branches with gt70 bootstrap support

19
Comparison Results
20
Results

• Out of the four comparisons, the JC corrected
  model had the most shared branches
• For one comparison, the K2P model had the most
  shared branches
• The parsimony/mismatch comparison fared the
  poorest

21
Conclusions

• In the NJ-ML consensus comparison, the JC
  corrected model scored best in 3 of 4 trials
• Of the 3 ML trees created for each model of
  evolution, the JC corrected model was given the
  highest ML score
• The ML consensus JC corrected model produces a
  topology most like the one we would expect based
  on biological intuition
• We therefore conclude that the JC corrected model
  of evolution best fits our data set

22
References

• 1 Krane, D.E. and Raymer, M.L. Fundamental
  Concepts of Bioinformatics. 2002. (ISBN
  0-8053-4633-3)
• 2 Lio, P., Goldman, N. Models of Molecular
  Evolution and Phylogeny. Genome Research.
  81233-1224, 1998.
• 3 Mount, D. W. Bioinformatics Sequence and
  Genome Analysis, 2nd ed., 2004. (ISBN
  0-87969-712-1)
• 4 Phylip Program, version 3.62. URL
  http//evolution.genetics.washington.edu/phylip/
• 5 PhyloDraw Program, version 0.8. URL
  http//pearl.cs.pusan.ac.kr/phylodraw/
• 6 Saitou, N., and Nei, M. The neighbor-joining
  method A new method for reconstructing
  phylogenetic trees. Mol. Biol. Evol. 4406-425,
  1987.
• 7 Subtree Pruning and Regrafting. URL
  http//www.hyphy.org/docs/analyses/methods/spr.htm
  l