Hi there,

I'm interested in comparing dN/dS ratios for concatenated alignments of all 13 protein coding sequences from 49 fish mitochondrial genomes.

I have two groups of fish and I want to know if the dN/dS ratios are significantly different from each other.

I'm thinking of using a phylogenetic t test Multimedia File Viewing and Clickable Links are available for Registered Members only!!  You need to Login as it's designed for use with continuous traits.

What are your thoughts? Is there a more robust or applicable test available? Many papers seem to use Fisher's exact tests or Mann-Whitney tests but there are reasons to suspect that they are ineffective for comparative data in a phylogenetic context.

Many thanks

-Jeff

Hi Jeff,

I wouldn't recommend a phylogenetically corrected t-test. Instead, I've adapted some HyPhy code that will set up a model where each branch is a random-effects mixture of 3 categories: two <=1 and one >1. You are able to divide the branches up into foreground, background, and other, by tagging the relevant branches in the phylogeny with {FG} or {BG} tags. "Other" branches do not need explicit tags, and are not included in any comparisons, but they do get their own set of parameters so they don't interfere with the BG/FG comparisons. An LRT compares FG to BG branches for differences in their selection parameters (the three omegas and two weights) where the null assumes they share all parameters, and the alt lets their parameters differ. Note that *any* differences in selection parameters will be identified (the big omega may increase while the small one decreases, so, even though the mean omega stays constant, a difference will be identified.)

You can find the code and example files here: Multimedia File Viewing and Clickable Links are available for Registered Members only!!  You need to Login

The bundled example is based on the rhodopsin dataset (It is quite slow - sorry!), comparing deep sea fish against surface animals, with the connecting branch untagged. The most convenient tree tagging strategy I've found is to color the branches in figtree (see coloredTree.nex), then find-and-replace color tags with {FG} or {BG}, and save the newick portion of the tree (see miniRhod.tre). The method had the preliminary acronym "AWEDIEBL", but I forgot what it stands for.

Let me know if you have any problems with it!
Ben Murrell

Many thanks Ben!

I'll take a crack at it tonight

Hi Ben,

Is there a functional difference between the tag names? i.e. does it matter which group I label with {FG} or {BG}

They sound like abbreviations of foreground/ background... that's why I'm double checking.

Also, is there a published paper out there that uses this method/ provides a detailed description? Your explanation was good enough for me but I'd like something more complex to cite for keen reviewers.

Cheers

-Jeff

Hi Jeff,

For this test there should be no functional difference. The paper describing this (and a few other tests) is in preparation, but the bulk of the methodology is described in Multimedia File Viewing and Clickable Links are available for Registered Members only!!  You need to Login

All Ben's test does is "instantiate" the class of branch-site type models to share some parameters among groups of branches.

Sergei

Thank you for taking the time to help me! If you guys ever find yourself in Guelph, Ontario I owe you a beer or two

I just remembered what the acronym stands for:
"Alignment Wide Evidence for Differences In Evolution Between Lineages".