How can the node distances affect to the omega values on the site model performed by codeml?

Question

[DR]

Hi! I'm performing a site positive selection test with codeml. I used the p-distance for obtaining the tree topology that then I used in order to contrast the different models for detecting positive selection. In spite of obtaining the same tree typology, could you give me an idea about how can the underestimated distance values affect to the dN/dS estimates?

Thanks!!

Answer 1

[Brice Sarver]

I used the p-distance for obtaining the tree topology that then I used in order to contrast the different models for detecting positive selection.

If you are fixing the tree (i.e., not letting codeml optimize topology/branch lengths), I would recommend using a more sophisticated means of estimating the tree. I would expect to see, at the very least, corrected p-distances if I was a reviewer. Likelihood-based approaches are better. Remember to transform your tree from units of substitutions/site to substitutions/codon.

Think of the branch lengths as your data for these analyses. The longer the branch lengths among your taxa, the more power you will have to infer positive selection; this is why it's hard to confidently infer positive selection acting on codons in closely-related lineages. Therefore, if your branch lengths are problematic, you might be drawing incorrect conclusions.