Hi Sergei,

In the MEME plos-genetics article it is written that :

“although MEME is considerably more powerful than existing methods at detecting bursts of selection, it still requires that a measurable proportion of lineages (5-10%) experience non-synonymous evolution at a site”

Let as assume we have ~10 orthologous sequences in an alignment. Also, let as assume that positive selection events occurred in only in one single branch of this tree, for only 5% of the sites. Is MEME designed for detection under such extreme scenario?
Under such scenario, will it be better to use branch-site REL (BranchSiteREL.bf), despite its inability to detect specific sites (as it only detects their proportions) ?

Best wishes,
Assi

Hi Assi,

In your instance, BS-REL would be the better choice to test for selection along a fixed lineage.
You could use MEME to explore which sites along a given lineage could have been affected by selection, but as the MEME paper says none of the methods can really drill down to the level of a single site/single branch with statistical confidence. That is because very little can be said based on the sample size of essentially one (a single realization of the evolutionary process), unless you take other information into account: allele frequency change, etc.

Sergei

Thanks Sergei

What is the minimal number of sequences for branch-site-REL and MEME ?  (I'm just running BranchSiteREL.bf with 5-8 sequences, and  see quite a lot of indications for episode selection, so I hope this small amount of sequences is enough, and also for MEME)

Also : should I give HYPHY trees created using the input alignments, or based on official standard species tree ?
(I’m asking this because tree branch-lengths may be program dependent, and also because short alignments,say 200 codons, tend to give wrong species topology)

Assaf

Hi Assaf,

In the scenario you describe (just one lineage under selection), it would be better to use BS-REL to test for selection on that particular lineage, sacrificing the site-by-site resolution. MEME can return significant results based on a single branch (e.g. if the branch is really short), but I would expect there to be fairly little power in a 10-sequence alignment.

Sergei

Thanks a lot Sergei,

I understand the above limitation ( my idea was to use MEME to expand my search to cases were selection events occurred at least in one particular specific branch)

just to understand it correctly:
1) About input trees – should they always be calculated from the corresponding input alignment?
3) about alignments with gap: should the region with gaps be removed ?  ( in such cases I see much more indications for selection in BS REL)

Best,
Assaf

Hi Assaf,

1). Yes, it is best to use the trees inferred from the sequences, probably with a nice ML-based program like Raxml or PhyML, before you upload the files to Datamonkey or feed them to HyPhy
2). The treatment of gaps is up to you: if you suspect the quality of the alignment is poor (some programs, like HMMER actually report it for you), then it is best to chuck gappy sites. If only a few sequences have gaps and the rest are well-aligned at a site, then by discarding this site, you will be throwing away good data.

Sergei


as wrote on Aug 13^th, 2013 at 10:07am:

Thanks a lot Sergei, I understand the above limitation ( my idea was to use MEME to expand my search to cases were selection events occurred at least in one particular specific branch) just to understand it correctly: 1) About input trees – should they always be calculated from the corresponding input alignment? 3) about alignments with gap: should the region with gaps be removed ?  ( in such cases I see much more indications for selection in BS REL) Best, Assaf