Tom Doak writes:
>working in a ciliated protozoa that is quite GC poor, and so we expect
>there to be a bias in the freq of different mutations, which led me to
>ask about transition/transversions, but of course that's not how one gets
>to be GC poor, or rather, just a two parameter model isn't enough, as I
>understand it. So is the answer that we just not worry about it?
>By the way, what we are currently doing is not so much tree building, but
>calculations of divergence and synonymous vs nonsynonymous to try to show
>conservation of a coding region.
The genetic code sets up to correlate transitions with synonymous
substitutions, so it would be best to incorporate a correction for
that component of the bias. Other componenets don't have a strong
correlation with the code (unless your protein is full of met or
something like that) and so it's not obvious that any correction
is relevant. Also, if you're comparing two extant sequences, then
the directional information is lost anyhow. ie. if you have a bias for
A->G over G->A and your sequences differ by an A vs. G, do you up
weight it or down weight it?
One real simple control is to recalculate out of frame. I'd be real
nervous if your computer program said there was conservation of
nonsynonymous positions in all three reading frames.