Welcome, Guest. Please Login
YaBB - Yet another Bulletin Board
 
  HomeHelpSearchLogin  
 
PARRIS and BS-REL (Read 3002 times)
perry
YaBB Newbies
*
Offline


Curious HyPhy user

Posts: 4
italy
Gender: female
PARRIS and BS-REL
May 14th, 2012 at 3:43am
 
Hi,
I have a couple of probably naive questions. I'm analysing a group of
functionally related gene (I have at least 24 species/gene) with both
PAML and Datamonkey.

First question concerns PARRIS: for some genes I get significant
results with PAML (both M1a/M2a and M7/M8) and there is no evidence of
recombination (based on GARD); yet, the PARRIS analysis indicates
there is no evidence of positive selection. Still, if I run MEME or
REL I find many sites with evidence of recurrent positive selection
(several of which correspond to M8 BEB). So, how do I interpret the PARRIS
results?

Second question is about branch-site REL: whether or not I find
evidence of positive selection for specific lineages, I often get
phylogenies that poorly reflect the known phylogeny of mammals, or at
least some species seem to be totally misplaced (I use the same
alignments with DNAml and I  get the expected mammalian tree). Is this
expected to happen? and how do I interpret these results?

Thanks a lot for your help, you provide a great service!
Perry
Back to top
 
 
IP Logged
 
konrad
Junior Member
**
Offline


I love YaBB 1G - SP1!

Posts: 53
Re: PARRIS and BS-REL
Reply #1 - May 14th, 2012 at 4:45pm
 
Dear Perry,

Re interpretation of the PARRIS results: with no evidence of recombination, PARRIS essentially reduces to an M1a/M2a analysis except that synonymous rate variation is taken into account. This leads (on the one hand) to a loss of power while (on the other) safeguarding against false positives that may result when PAML neglects actual synonymous rate variation (which has been demonstrated to be common). In general (for the implementations I know about), HyPhy implementations take synonymous rate variation into account while PAML implementations do not.

That said, we now recommend MEME as it is the only one of these methods which takes heterotachy into account - this typically leads to an increase in power.

Re phylogeny inference with branch-site REL: I am not sure exactly which model HyPhy uses to infer the phylogeny in this case, but in general our codon model implementations are not designed for inferring phylogenies - in particular, there is no built-in safeguard against overparameterization (this is built into the hypothesis test for selection, but not into phylogeny inference). If inferring a phylogeny under a codon model, it would be better to first run a model selection step to check that data set is informative enough to properly infer the model in question. Instead of this, we usually just infer phylogenies under a simpler (nucleotide) model; even when inferring under a codon model is justified, in most cases we don't expect it to make enough difference to justify the computational cost.

In cases such as mammals where the phylogeny is known you should definitely input that phylogeny rather than try to infer it from your data set - the known phylogeny is based on far more information than just the sequences in your particular data set.

I hope this helps,
Konrad Scheffler
Back to top
 
WWW WWW  
IP Logged
 
perry
YaBB Newbies
*
Offline


Curious HyPhy user

Posts: 4
italy
Gender: female
Re: PARRIS and BS-REL
Reply #2 - May 15th, 2012 at 8:25am
 
yes, it helps!
Thank you so much
perry
Back to top
 
 
IP Logged