The model permits an analytical computation of the marginal likelihood of an alignment given a phylogenetic tree, thereby offering an evaluation of how the estimated model generalizes to novel observations.
It became popular because of its analytic tractability (cf. Siddharthan et al., 2005), and it is simple enough so that the marginal likelihood of an alignment, given a phylogenetic tree can be computed analytically.
This scoring function measures the sequence-template similarity using a set of regression trees, which take as input protein features and output the log-likelihood of an alignment state (i.e. match or gap).
Conversely, ML scores can also be used to judge the likelihood of a multiple alignment reflecting sequence evolution.
This matrix can then be used to calculate the likelihood of a sequence alignment under a fixed tree topology, using established procedures (Felsenstein 1981; Yang 2006).
Such a substitution model would be able to calculate the likelihood of a singlet alignment while making use of codon frequencies and differentiating between synonymous and non-synonymous changes.
Maximum parsimony, neighbor joining and maximum likelihood analyses of an alignment of 50 concatenated protein sequences including a total of 19,836 aligned sites (Additional file 2), all yielded identical fully resolved topologies with high bootstrap support.
DOI: http://dx.doi.org/10.7554/eLife.07335.012 10.7554/eLiFigure35.013 Figure 2 figure supplement 2. Maximum likelihood phylogeny of an alignment of the concatenated nucleotide sequences the enterobactin locus of 13 genes (34,732 NT; containing entABCDEFS, fepABCDG and fes ).
Scoring a site using a PMM involves computing a log-likelihood ratio of an alignment column σ of N sequences given evolutionary models of the motif and background θM and θB, respectively.
The likelihood of a multiple sequence alignment D = { d1,..., d n } of n sites given their phylogenetic tree T and the model Q is (1) where L T, Q| d i ) is the likelihood of site d i given tree T and model Q that can be efficiently calculated by a pruning algorithm [ 29].
