Building phylogenies

Built-in Phylogenetic reconstruction

By distance method

Given an alignment, a phylogenetic tree can be generated based on the pair-wise distance matrix computed from the alignment.

Fast pairwise distance estimation

For a limited number of evolutionary models a fast implementation is available. Here we use the Tamura and Nei 1993 model.

>>> from cogent import LoadSeqs, DNA
>>> from cogent.evolve.pairwise_distance import TN93Pair
>>> aln = LoadSeqs('data/primate_brca1.fasta')
>>> dist_calc = TN93Pair(DNA, alignment=aln)
>>> dist_calc.run()

We can obtain the distances as a dict for direct usage in phylogenetic reconstruction

>>> dists = dist_calc.getPairwiseDistances()

or as a table for display / saving

>>> print dist_calc.Dists[:4,:4] # truncated to fit screens
Pairwise Distances
============================================
Seq1 \ Seq2    Galago    HowlerMon    Rhesus
--------------------------------------------
     Galago         *       0.2157    0.1962
  HowlerMon    0.2157            *    0.0736
     Rhesus    0.1962       0.0736         *
  Orangutan    0.1944       0.0719    0.0411
--------------------------------------------

Other statistics are also available, such the as the standard errors of the estimates.

>>> print dist_calc.StdErr[:4,:4] # truncated to fit screens
Standard Error of Pairwise Distances
============================================
Seq1 \ Seq2    Galago    HowlerMon    Rhesus
--------------------------------------------
     Galago         *       0.0103    0.0096
  HowlerMon    0.0103            *    0.0054
     Rhesus    0.0096       0.0054         *
  Orangutan    0.0095       0.0053    0.0039
--------------------------------------------

More general estimation of pairwise distances

The standard cogent likelihood function can also be used to estimate distances. Because these require numerical optimisation they can be significantly slower than the fast estimation approach above.

>>> from cogent import LoadSeqs, DNA
>>> from cogent.phylo import distance
>>> from cogent.evolve.models import F81
>>> aln = LoadSeqs('data/primate_brca1.fasta')
>>> d = distance.EstimateDistances(aln, submodel=F81())
>>> d.run()

The example above will use the F81 nucleotide substitution model and run the distance.EstimateDistances() method with the default options for the optimiser. To configure the optimiser a dictionary of optimisation options can be passed onto the run command. The example below configures the Powell optimiser to run a maximum of 10000 evaluations, with a maximum of 5 restarts (a total of 5 x 10000 = 50000 evaluations).

>>> dist_opt_args = dict(max_restarts=5, max_evaluations=10000)
>>> d.run(dist_opt_args=dist_opt_args)
>>> print d
============================================================================================
Seq1 \ Seq2    Galago    HowlerMon    Rhesus    Orangutan    Gorilla     Human    Chimpanzee
--------------------------------------------------------------------------------------------
     Galago         *       0.2112    0.1930       0.1915     0.1891    0.1934        0.1892
  HowlerMon    0.2112            *    0.0729       0.0713     0.0693    0.0729        0.0697
     Rhesus    0.1930       0.0729         *       0.0410     0.0391    0.0421        0.0395
  Orangutan    0.1915       0.0713    0.0410            *     0.0136    0.0173        0.0140
    Gorilla    0.1891       0.0693    0.0391       0.0136          *    0.0086        0.0054
      Human    0.1934       0.0729    0.0421       0.0173     0.0086         *        0.0089
 Chimpanzee    0.1892       0.0697    0.0395       0.0140     0.0054    0.0089             *
--------------------------------------------------------------------------------------------

Building A Phylogenetic Tree From Pairwise Distances

Phylogenetic Trees can be built by using the neighbour joining algorithm by providing a dictionary of pairwise distances. This dictionary can be obtained either from the output of distance.EstimateDistances()

>>> from cogent.phylo import nj
>>> njtree = nj.nj(d.getPairwiseDistances())
>>> njtree = njtree.balanced()
>>> print njtree.asciiArt()
                    /-Rhesus
          /edge.1--|
         |         |          /-Galago
         |          \edge.0--|
         |                    \-HowlerMon
-root----|
         |--Orangutan
         |
         |          /-Human
          \edge.2--|
                   |          /-Gorilla
                    \edge.3--|
                              \-Chimpanzee

Or created manually as shown below.

>>> dists = {('a', 'b'): 2.7, ('c', 'b'): 2.33, ('c', 'a'): 0.73}
>>> njtree2 = nj.nj(dists)
>>> print njtree2.asciiArt()
          /-a
         |
-root----|--b
         |
          \-c

By least-squares

We illustrate the phylogeny reconstruction by least-squares using the F81 substitution model. We use the advanced-stepwise addition algorithm to search tree space. Here a is the number of taxa to exhaustively evaluate all possible phylogenies for. Successive taxa will are added to the top k trees (measured by the least-squares metric) and k trees are kept at each iteration.

>>> import cPickle
>>> from cogent.phylo.least_squares import WLS
>>> dists = cPickle.load(open('data/dists_for_phylo.pickle'))
>>> ls = WLS(dists)
>>> stat, tree = ls.trex(a = 5, k = 5, show_progress = False)

Other optional arguments that can be passed to the trex method are: return_all, whether the k best trees at the final step are returned as a ScoredTreeCollection object; order, a series of tip names whose order defines the sequence in which tips will be added during tree building (this allows the user to randomise the input order).

By ML

We illustrate the phylogeny reconstruction using maximum-likelihood using the F81 substitution model. We use the advanced-stepwise addition algorithm to search tree space, setting

>>> from cogent import LoadSeqs, DNA
>>> from cogent.phylo.maximum_likelihood import ML
>>> from cogent.evolve.models import F81
>>> aln = LoadSeqs('data/primate_brca1.fasta')
>>> ml = ML(F81(), aln)

The ML object also has the trex method and this can be used in the same way as for above, i.e. ml.trex(). We don’t do that here because this is a very slow method for phylogenetic reconstruction.

Building phylogenies with 3rd-party apps such as FastTree or RAxML

A thorough description is Using phylogeny application controllers to construct phylogenetic trees from alignments.