M1a vs M2a convergence issues

[Emily Kopania]

Hello,

I am testing for positive selection on many genes using the M1a vs M2a site model, and I am running into convergence issues. 

First, 3152 out of 11748 genes are giving me a "check convergence" message in the codeml output file. 

Second, 6472 genes out of 8341 that did not have the "check convergence" message have a negative LRT score. 6083 of these are between -1 and 0, which I have had happen with other datasets and have estimated the LRT to be 0, but this still seems like a lot of loci to have an LRT < 0. Do you know if this is cause for concern, or what the reasons may be for this lack of convergence? 

I have attached example output, control, alignment, and tree files. ENSMUSP00000000028 gave a "check convergence" message whereas ENSMUSP00000000010 did not. Both gave slightly negative LRT scores.

Thank you!

[Sandra AC]

Hi there,

If you use the search bar in the google bar and type the keywords "check convergence", you can find various posts that can help answer your question. E.g., in this post, Ziheng pointed out that "if different runs produce the same thing, ignore the warning message".

Hope this helps!
S.

[Emily Kopania]

Okay, thanks!