Dear Sergei,

I have programmed Felsenstein's pruning algorithm as an exercise in Python and get the same likelihood for a tree with fixed parameters as HyPhy, except when the codon/nucleotide at a site is unknown for one or more sequences.

I was just wondering how HyPhy handles such unknown nucleotides/codons in the initialization of the pruning algorithm. For example, if the nucleotide at a site is, say, cytosine and sites are ordered [A,C,G,T], then the conditional likelihoods at the tip of the tree are initialized as [0,1,0,0] in the pruning algorithm. When the nucleotide is unknown, is the initialization [1,1,1,1] or [0.25,0.25,0.25,0.25]? Although, you should arrive at the same MLEs either way, the maximum value of the likelihood function will differ. What does HyPhy do?

Thanks in advance,
Miguel

Dear Miguel,

Hyphy maps an unknown (or a gap) to [1,...,1,1] - this is equivalent to simply removing the branch terminating in the n-fold ambiguity at that site.

Cheers,
Sergei