HyPhy message board - Print Page

HyPhy message board
http://www.hyphy.org/cgi-bin/hyphy_forums/YaBB.pl Methodology Questions >> How to >> branch-specific BS-REL? http://www.hyphy.org/cgi-bin/hyphy_forums/YaBB.pl?num=1351631333 Message started by Vinny Lynch on Oct 30^th, 2012 at 2:08pm

Title: branch-specific BS-REL?
Post by Vinny Lynch on Oct 30^th, 2012 at 2:08pm

Hi All,

Is there a way to restrict a BS-REL analysis to a particular lineage? I'm not really interested in testing all branches since there is only one I am interested in. I tried looking at the .bf file for BS-REL but it was not obvious to me where I might change something.

Thanks
Vinny

Title: Re: branch-specific BS-REL?
Post by Vinny Lynch on Oct 31^st, 2012 at 1:06pm

It just occurred to me that targeting a specific lineage for the BS-REL test a priori will not produce different results from running the model on the full dataset (other than saving computational cost and time). So might it be statistically OK to disregard the other lineages and assess the evidence for selection from the LRT and p-value of only my a priori selected lineage (rather than the p_Holm)?

Title: Re: branch-specific BS-REL?
Post by Sergei on Nov 5^th, 2012 at 3:34pm

Hi Vinny,

This is correct. I am also attaching a file that permits you to select lineages for analysis.

Sergei

http://www.hyphy.org/cgi-bin/hyphy_forums/YaBB.pl?action=downloadfile;file=BranchSiteRELMarked_002.bf (15 KB | 11 )

Title: Re: branch-specific BS-REL?
Post by Vinny Lynch on Nov 7^th, 2012 at 8:19am

Hi Sergei,

Thanks for the file, that will certainly help and save your datamonkey servers from analyzing my dataset!

I'm curious which case you find more convincing from these two nested BS-REL analysis.

A dataset of 51 mammalian species with 213 codons where BS-REL identifies positive selection in human with mean omega=10, omega1=0.0008 (P1=0.68), omega2=0.0008 (P2=0.56), and omega3=1000 (p3=0.09) at LRT=5.22 and p-value=0.01.

Or dataset of 18 primate species selected from the larger 51 species dataset where BS-REL identifies positive selection in human with mean omega=10, omega1=5E-9 (P1=0.88), omega2=7.5E-8 (P2=0.03), and omega3=1000 (p3=0.085) at LRT=8.7 and p-value=0.0017.

Both are statistically signifiant and have similar results, but should one always prefer a larger dataset (so long as the alignment is clear)?

Thanks,
Vinny

Title: Re: branch-specific BS-REL?
Post by Sergei on Nov 7^th, 2012 at 10:38am

Hi Vinny,

I would report both (except in your first example the probabilities add up to > 1); the consclusions are the same: a small proportion of sites (about 9%) is under very strong diversifying selection. Why the different LRT?

1). The analysis may not have fully converged for the alternative model with 51 species.
2). The lineages you added for the large data set may have had enough of an effect on the inference for the human lineage.

Not too much of a concern though.

Sergei