I received this question by e-mail from Tim Schmidt.

Question: I'm running this analysis (data set attached, vert mtDNA code). It seems to work fine for the SLAC and ARS analysis, but for the maximum likelihood analysis the detailed results show this:

1 0.973591 0.0207577 0.0215583 -0.885854 0 nan nan nan nan
2 0.994573 0.172984 0.1754 -0.763834 0 nan nan nan nan
3 1.00244 0.205175 0.206158 -0.74122 0 nan nan nan nan
4 1.00413 0.166192 0.166661 -0.779037 0 nan nan nan nan

What does the nan refer to?


'nan' is a an abbereviation of Not A Number. The 'nan' will occur within full likelihood results if the inferred distribution of dN-dS is entirely below 0 or entirely above 0. In this case all Bayes factors for the event dN>dS (or dN<dS) are undefined because the odds for one of the events are infinite. Indeed, in your case dN is inferred to be always less than dS, and the prior AND posterior odds for dN>dS are therefore both 0, since prior (and posterior) probability of dN>dS is 0.

In other words, if all values of dN-dS are on the same side of '0', then the method suggests that all sites are either positively (dN-dS>0) or negatively (dN-dS<0) selected. This will usually happen for small datasets with little site-to-site rate variability.

HTH,
Sergei