I have been using the relative rates test on a set of Insect species from 6 orders with the main interest being the evolutionary rate of Hymenopterans compared to the others. I have done the test for full mtDNA protein-coding sequences, for mtDNA rRNAs, for 18SrRNA, an Argk gene, a Pepck gene with great success.

We are having a problem with the so-called wingless gene, which admittedly has a very wacky tree that has lots of polytomies. The tests run beautifully with p-values coming out just as I would expect on the basis of the ML branch lengths. The problem is that the branch lengths to the MRCA can be huge for this gene. Really unrealistic values in the millions.

So I wonder if there is some kind of computational problem that involves the polytomies even though I do not see a pattern that corresponds with that fact.

We have a paper that is close to being accepted but I am uncomfortable with this particular test.

BTW, I also can not get the relative rate test Cadence to work with this gene.

Hi pscuu,

It is entirely possible for a statistical test (e.g. the likelihood ratio test) to perform well even though the point estimates of the parameters (branch lengths) are not precise. I bet in your case, these long branches have very high variance associated with the branch length, e.g. the confidence intervals on the estimates are wide. It sounds like your gene simply may have saturated substitutions along some lineages; we see that a lot in viruses, for example (where evolutionary rates are very high). I don't know the first thing about insect phylogenies, but I would imagine the same intuition applies.

Sergei