HyPhy message board - Print Page

HyPhy message board
http://www.hyphy.org/cgi-bin/hyphy_forums/YaBB.pl Datamonkey Server >> Datamonkey feedback >> Strange REL results http://www.hyphy.org/cgi-bin/hyphy_forums/YaBB.pl?num=1233925362 Message started by Hiacynt on Feb 6^th, 2009 at 5:02am

Title: Strange REL results
Post by Hiacynt on Feb 6^th, 2009 at 5:02am

Dear Sergei,
I'm running a REL analysis using HKY85 on a clade of sequences, and I'm getting unusual results, namely all positions have EXACTLY the same values of estimates. While the sequences do share a certain degree of similarity, they are not 100% identical, which is a situation that I would suspect to produce such results. Could you help me out with what might cause this, or how should I treat this?
Just for referencem I have run the analysis several times, and the results persisted, and that's why I came here.

Title: Re: Strange REL results
Post by Sergei on Feb 6^th, 2009 at 9:38am

Dear Hiacynt,

This is indeed strange; could you please post a link to the datamonkey results page?

Cheers,
Sergei

Title: Re: Strange REL results
Post by Hiacynt on Feb 7^th, 2009 at 4:09am

Here:
datamonkey.org/spool/upload.563270981528200.1_rel.php

(the board system won't let me post an active link before I have 5 posts, so I had to truncate it)

Title: Re: Strange REL results
Post by Sergei on Feb 7^th, 2009 at 10:18am

Dear Hiacynt,

Because your alignment has only 6 sequences, the REL method found only two rate classes (effectively) in your data -- both with dN<dS (i.e. purifying selection). When this happens, empirical Bayesian inference of positive selection is impossible (since there is no evidence for dN>dS) and the method will output per-site results that are all identical (0 or 1 posterior/prior probability).

This is one of the reasons that inferring selection on small alignments is a difficult and statistically challenging undertaking.

HTH,
Sergei