Hi,

I am wondering which method is best to look for positive selection in a comparative transcriptome dataset with very few species and very low sequence divergence.

I have a set of codon alignments for several thousand genes from 4 ingroup species, one of which is represented by two different varieties for a total of 5 ingroup terminals.  There is a relatively far outgroup that we've added to an equivalent set of codon alignments, so in those alignments we have 6 terminals.

I'm trying to decide whether to run PARRIS (without partioning), because of the very small size and low sequence divergence in the data, or SLAC and FEL with a high nominal alpha level (say 0.25 or 0.3).  Maybe all of the above is best?

Given time limitations on this study (i.e. we need to get it done within the next month) I was not planning to use REL.  Currently we have ~14,000 alignments, although I may filter these down with more stringent 1-to-1 orthology assignments and dropping all alignments with less than 5 or 10 SNPs.

Any feedback would be greatly appreciated.
Thanks,
Dan.

Hi Sergei,

Thanks again for the fast response.  I'll read the fingerprinting paper.

One last question, would PARRIS and the fingerprinting run faster in MP or MPI?

Thanks,
Dan.

Hi Sergei,

If there's an issue I could code it without using genomefitters, but it does seem like that pipeline is well-suited for my needs.

If you can, please let me know if it's MPI enabled.

Thanks,
Dan.

Hi Sergei,

YES, we are really strapped for time (it's a long story...) and we would greatly appreciate being able to run the analyses on your cluster.

By the way, I was really blown away by the fingerprinting paper!  I really liked the efficiency of parameterization of the GB models, as well as the heuristics for optimizing the number of rate classes and approximating the ESD posterior using SIR.  It made me wonder if it's worth it to recode a bunch of "standard" REL-type dN/dS methods with GB models; my naive interpretation is that, apart from other potential benefits, these may have better power with small datasets due to the lower number of parameters.

I'm really eager to apply the fingerprinting method to our data.  As a matter of fact, we were already going to look for correlations between gene expression clustering (we have species and tissue specific data) and positive selection.  The fingerprinting gives us a more meaningful evolutionary measure for comparisons!  It would be great to see if genes with similar tissue or species variation have or don't have similar evolutionary fingerprints.

We already have codon alignments for our ingroup taxa, and we will have the alignments with the outgroup (i.e. 6th) sequence very soon.  It may be worth running the jobs with the ingroup-only data in the mean time and redo them later with all 6 taxa, in part because we'll be throwing out a bunch of genes that don't have a clear 1:1 ortholog in the outgroup taxon.  I know, there are perils to running these analyses with yet one fewer sequence.

Thanks so much for your offer regarding the cluster!
-Dan.