Hello!

I'm trying to determine if a group of sequences is evolving under positive selection. I am interested in sites more than branches. I have addressed two approaches. One is using codeml from PAML and the other one using MEME.

Both approaches give me evidence that YES, there is positive selection. I used LRT between M7 and M8 in CODEML and then the Bayes Empirical Bayes (BEB) to calculate the posterior probabilities that each codon is from a site class of positive selection under model M8.

Now, the problem I have is that MEME and BEB give me completely different results regarding to which sites are the ones evolving under positive selection.

I know that the approaches of both methods to identify the sites are differents, but I was wondering if someone could give a possible explanation for this discrepancy.

Thanks a lot!

Gabriel

HI Gabriel,

There are THREE major differences between M8+BEB and MEME

1). MEME allows dS to vary from site to site. M8+BEB does not (see Multimedia File Viewing and Clickable Links are available for Registered Members only!!  You need to Login for examples of how this can cause M8+BEB to confuse relaxation of constraint for positive selection).
2). M8+BEB is a random effects method, while MEME is a fixed effects method when it comes to estimating dS and dN at individual site. MEME will have more flexibility to fit the distribution of dS and dN (unlike M8+BEB which is limited to, say, beta + w)
3). Most critically, MEME can find episodes of selection (a burst of high omega values followed by negative selection), but M8+BEB can find only evidence of average selection (over branches). The MEME manuscript, where this is explained in detail, is currently in revision, but the following picture should clarify the difference.

Sergei

---

Dear Sergei,

Thank you very much for your quick and teaching reply.

Ok, so I understand that sites identified by MEME and not by M8+BEB can be explained by your points 2) and 3).

But, for example, sites identified by M8+BEB as evolving under positive selection and not identified by MEME could be then explained by a relaxation of constraint rather than positive selection? I guess MEME can also find episodes of selection (as you say, a burst of high omega values followed by negative selection), as well as it should identify evidence of average selection over branches, isn’t it?

Thank you very much again.
Cheers, Gabriel

Hi Gabriel,

I would say that sites that were found by M8+BEB but not by MEME could be due to

[olist]

Relaxed selection -- this is especially likely if the site has a lot of synonymous and non-synonymous diversity

Statistical artifacts in either method (there are many possibilities).
[/olist]

Can you provide an example of such site (e.g. with substitutions mapped to branches using MEME/datamonkey)?

Sergei

Thanks Sergei,

The results from MEME are here: Multimedia File Viewing and Clickable Links are available for Registered Members only!!  You need to Login

For this alignment, M8+BEB identified sites 4, 260 and 285 with Pr(w>1) > 0.9.

Gabriel