Alternative Likelihood searching algorithms

1
Alternative Likelihood searching algorithms

• Quantitative Phylogenetics
• Spring 2008

2
Outline

• PhyML
• GARLI
• RaxML

3
PhyML

• simple hill-climbing algorithm that adjusts tree
  topology and branch lengths simultaneously
• Starts with a distance-based tree and improves on
  it
• Can optimize model parameters
• Based on computer simulations, it is at least
  equivalent to other ML programs in terms of
  topological accuracy and maximum likelihood
  optimization, but as fast as parsimony and
  distance-based approaches
• Web-based and executables (not appropriate for
  parallel processing)
• Cannot handle multiple data partitions
• http//atgc.lirmm.fr/phyml/
• Guindon Gascuel 2003

4
Genetic Algorithms

• Heuristic approach that is a practical
  application of natural selection
• Population of individuals
• Each individual is characterized by a Fitness
  function
• Higher fitness is better solution
• Based on their fitness, parents are selected to
  reproduce offspring for a new generation
• Fitter individuals have more chance to reproduce
• New generation has same size as old generation
  old generation dies
• Offspring has combination of properties of two
  parents and new mutations
• If well designed, population will converge to
  optimal solution

5
Genetic Algorithms in Phylogenetics
Individual
Chromosome
-lnL score
1
best
1027
2
1038
1039
3
4
1055
worst
5
1061
Population of five individuals
6
Genetic Algorithms in Phylogenetics
Spin the wheel five times to determine next
generation (population size constant)
Each individual (tree/branch-lengths/parameters)
gets a slice proportional to its fitness
(likelihood)

• Best individuals leave more offspring
• Mutations mostly in the form of branch
  rearrangements
• Usually, best individual is protected from
  mutation

7
Genetic Algorithm for Rapid Likelihood Inference
(GARLI )

• Performs heuristic phylogenetic searches under
  the GTR model and its submodels, with or without
  gamma distributed rate heterogeneity and a
  proportion of invariant sites.
• Implementation is exactly equivalent to that in
  PAUP, so that likelihood scores obtained by each
  program are directly comparable.
• Interactive in Mac OS X GUI version other
  versions not interactive.
• The program reads non-interleaved Phylip and
  nexus datafiles.
• GARLI also implements a parallel MPI based
  algorithm for use on parallel computing clusters
  (note that the parallel version seeks to perform
  a more thorough tree search and does NOT reduce
  runtimes. It typically is only helpful on
  datasets with greater than about 500 sequences)
• Can optimize model parameters
• Cannot handle multiple data partitions
• http//www.bio.utexas.edu/faculty/antisense/garli/
  Garli.html
• Zwickl 2006 (Hillis lab)

8
Randomized A(x)ccelerated Maximum Likelihood
(RaxML)

• Performs heuristic phylogenetic searches
• starts by building an initial parsimony tree with
  dnapars from Felsensteins PHYLIP package
• standard subtree rearrangements by subsequently
  removing all possible subtrees from the currently
  best tree, and re-inserting them into neighboring
  branches up to a specified distance of nodes
  (inherited from fastDNAml).
• Implementation is NOT exactly equivalent to that
  in PAUP, so that likelihood scores obtained by
  each program are NOT directly comparable.
• Can optimize model parameters
• CAN handle multiple data partitions (including AA
  and DNA)
• Single processor and parallel versions
• CIPRES webserver
• http//8ball.sdsc.edu8889/cipres-web/Bootstrap.do
  
• http//icwww.epfl.ch/stamatak/index.htm

9
Statamakis et al. 2005
10
Statamakis et al. 2005
11

• Compared several programs
• Conclusions
• most of the current heuristic algorithms that are
  advertised as providing faster analyses than
  PAUP often do so at the expense of finding the
  optimal tree
• These algorithms are prone to getting stuck in
  local optima, even when the topography of the
  tree space appears to be relatively smooth
• Recommendation All fast programs should be
  used in conjunction with a more accurate program
  in order to guarantee that a good tree has been
  located.

12
References

• Guindon S, Gascuel O (2003) A simple, fast, and
  accurate algorithm to estimate large phylogenies
  by maximum likelihood. Syst Biol 52, 696-704.
• Morrison DA (2007) Increasing the efficiency of
  searches for the maximum likelihood tree in a
  phylogenetic analysis of up to 150 nucleotide
  sequences. Syst Biol 56, 988-1010.
• Stamatakis A (2006) RAxML-VI-HPC maximum
  likelihood-based phylogenetic analyses with
  thousands of taxa and mixed models.
  Bioinformatics 22, 2688-2690.
• Stamatakis A, Ludwig T, Meier H (2005) RAxML-III
  a fast program for maximum likelihood-based
  inference of large phylogenetic trees.
  Bioinformatics 21, 456-463.
• Zwickl, D. J., 2006. Genetic algorithm approaches
  for the phylogenetic analysis of large biological
  sequence datasets under the maximum likelihood
  criterion. Ph.D. dissertation, The University of
  Texas at Austin.