Open glitterseaworms opened 3 weeks ago
Hello, I was wondering if we should not consider codon sites that have significant p-values but have 0 branches listed? For example: Based on the likelihood ratio test, episodic diversifying selection has acted on 1 sites in this dataset (p≤0.05). | Codon | Partition | alpha | non-syn rate (beta) distribution, rates : weights | LRT | Episodic selection detected? | # branches | List of most common codon substitutions at this site |
---|---|---|---|---|---|---|---|---|
168 | 1 | 0.120 | 0.05/2808.48 : 0.00/1.00 | 15.009 | Yes, p = 0.0002 | 0 | [1]GTt>AAt,GTt>TAt,GTT>TCC,GTt>TCt,TCT>AAC,tcT>tcC |
Dear @glitterseaworms,
This is a bit counterintuitive, but here's an explanation. For the site you provide as an example, MEME
allocates 100% of the available ω weight to the positive selection regime: β = 2808.49.
When MEME calculates which, if any, branches may be subject to selection, it attempts to estimate a Bayes Factor (empirical Bayes), that β > α (or ω > 1) at that branch. However, the PRIOR probability for ω > 1 is estimated to be exactly 1, which makes the prior odds (needed to calculate Bayes Factors) infinite. So MEME
cannot calculate a Bayes Factor for sites where all the weight is assigned to β > α, and you end up with the 0
estimate.
Hope this helps, Sergei
Ah ok thanks for the explanation. In that case, would you say it is reasonable to hypothesize that the site may still be under diversifying selection then?
Dear @glitterseaworms,
Yes, and the selection is pervasive (applies to all branches), or at least we have no ability to resolve it further down to specific branches.
Best, Sergei
Thank you kindly for your help. I really appreciate you!