Hi HyPhy community,

I am trying to decide between running only PARRIS with 2 synonymous rate classes or also BranchSiteREL and MEME on a set of alignments with 6 taxa.

In general, there is a low level of sequence divergence in these alignments, which is why PARRIS was originally chosen for the positive selection analysis. The average dS is 0.01 among the ~11200 alignments in the dataset.

Given the additional information that can be gleaned from a "branch-site" approach and the reportedly increased power of the new methods BranchSiteREL and MEME, should I also run these new analyses on my genes or at least say on the top 20% most variable subset of them? Or, am I just not going to have enough information in my alignments to estimate the parameters of the more complex models in BranchSiteREL and MEME?

I realize BranchSiteREL and MEME are somewhat complementary to PARRIS, as they're designed to detect episodic as well as pervasive positive selection. However, the issue of lack of information relative to more complex models still applies.

So, what analyses make the most sense to run? Only PARRIS, or
PARRIS plus BranchSiteREL and MEME?

Thanks in advance!
Dan.

Hi Dan,

Well, PARRIS and MEME/BranchSiteREL look for different types of selection: the former detects only pervasive diversifying selection, while both of the latter methods can detect episodic and pervasive selection. I would run BranchSiteREL (assuming you have no recombination), because the test in this method is conceptually similar to PARRIS -- is there a non-zero proportion of sites under selection along any of the branches.

I would not worry about low sequence divergence too much here, all that would happen to BS-REL and MEME is that they would lose power. Low sequence divergence does not seem to lead to increased rates of false positives, if that's what you are worried about.

Sergei

Hey all!

Sorry for hijacking the thread, but my question is related to the initial question. I want to find which lineages are subject to episodic diversifying selection using BranchSiteREL, but I have a recombination breakpoint. Is there any workaround to be able to use it with recombinant sequences? Or do you have any recommendation on how proceed?

Thank you very much!

Hi Antoni,

There are two possibilities:

1). Split your alignment into non-recombinant fragments and analyze each through BranchSiteREL separately.
2). Remove recombinant sequences (if you know which ones there are).

Both are hacks, because the right thing to do is to identify shared lineages across the breakpoint (e.g. if you have a common subtree in both partitions), and model such lineages jointly across the partitions.

Sergei

Thanks, Sergei!  :)