Units Of Dn And Ds In Codeml Paml Output

Question

[jkn55]

I was just looking for some clarification on the meaning of the dn and ds outputs from a pairwise comparison in codeml of PAML.

Is dn the rate of non-synonymous subs per non-synonymous site? Or is it per codon?

Given they are rates, what is the unit of time? From reading the literature, I get the impression that the transfer matrix is normalised in such a way that the 'rate' is then the actual expected number of substitutions between the two sequences (per site type). Is this correct?

Further to this, if ds > 1 for a pairwise comparison, does this imply that all synonymous sites have expected to have been substituted?

Thanks for any help!

Answer 1

[jkn55]

Just to add information to this thread, I found this post on the PAML forum (with a reply from the author of PAML) that seems to suggest that dS is measured in number of synonymous changes per synonymous site. https://www.ucl.ac.uk/discussions/viewtopic.php?f=54&t=7594 Thanks.

[David W]

Hi JKN, sorry to have steered you in the wrong direction to start with and thanks for coming back to get the right answer recorded.

Answer 2

[David W]

JKN,

You're right that the rate is the expected number of substitutions, codeml is a bit odd in that the rate it gives is the expected number of _nucleotide_ substitutions per _codon_ (at least that's what you get for branch models, I presume the same applies to pariwise comparisons, since that would explain your dS > 1).

The PAML FAQ [PDF] has a little information on this in the section describing baseml.

EDIT: As JKN points out below the advise is wrong, for pairwise comparisons the value given is substitutions per site.

[jkn55]

Thanks for your reply and link to the info! Just another check, so from the documentation, I get that the mutation rates and branch lengths seem to be in these units...so how then do dN and dS relate to branch lengths? Based on what you said, this should be: (dN + dS) * numberofcodons for codeml? Thanks.

[hbw]

So how can we get dS>1 for pairwise comparisons?