Same log likelihood values for M1 and M2 models

[Minal Jamsandekar]

Dear All,

I am facing a problem with log likelihood values. I get exact values for M1 and M2 site models while nearly same values for M7 and M8 . I tried changing the parameters such as omega initial value, kappa initial value, number of Gamma categories etc. Also, I tried using the codeml.ctl file that is given in HIV NSsites examples. But all such attempts are futile. I get omega values and positively selected sites in result file. Only lnL values are causing problem.

Can anyone please suggest me a way out for this?

[Sungmin Lee]

Hi,

Did you check p associated to the omega? In case the p is zero, the omega value may not be so significant. 

Cheers,

[SP]

Hello,

I'm experiencing a similar issue. Following this example (https://embnet.vital-it.ch/CoursEMBnet/PagesPHYL07/Exercises/day2/day2.html), I ran the branch test and found a significant result for two different branches with the log likelihood tests. I followed up with the branch-sites tests to determine which sites were under selection in each branch, and there seemed to be some significant sites from the BEB test marked with **. To do the branch-sites test, I used the same marked tree and input files but changed the control file parameters for the branch-sites test.

When I compared log likelihoods (for results from both of the branches that had positive selection in the branch test prior), I noticed the log likelihoods (for null and alternate) were the exact same in both tests.

Do you know what this could mean?

Thank you!

[SP]

Hello!

I realized I should have included the following information in my earlier post:

In my tree, I marked multiple branches in a clade using #1 (the branches are nested).

Parameters for branch-sites test (in reference to the manual p. 38):

In the alternate model, I set control file parameters to model=2, NSsites=2, fix_omega=0, and omega=1

In the null model, I set model=2, NSsites=2, fix_omega=1, omega=1

Parameters for branch test (I ran this test first, before the branch-sites model):

Null: Model=0, NSsites=0, fix_omega=0, omega=1

*no branches labelled in tree

Alternate: Model=2,  NSsites=0, fix_omega=0, omega=1 (only changed Model from 0 to 2)

*branches labelled in tree

Is it possible I did not use the correct parameters for one of these tests?

Thank you!

[Ziheng]

your control-file settings for the branch-site models look OK to me or at least i can't see anything obviously wrong.  

it sounds like you have the same lnL for M1 and M2, so that the LRT is not significant, but the BEB calculation has identified sites under positive selection.

first run the program a few times to confirm that the results are stable between runs.

if computational issues are not a concern, then this may be a case of weak evidence for positive selection, so that the LRT and BEB give conflicting results.  did you get high posterior probabilities from the BEB calculation?

your control-file settings also refer to the branch model.  but for the branch model there is no BEB calculation.  

the branch model assumes all sites have the same w for any branch so the estimate of w for that branch is a kind of average over all sites.  this is often less than 1, and the test is a very conservative way of testing for positive selection on branches.       

ziheng