Bonferroni correction

[hvk]

Hi PAML Users,

I have run the site model test M1-M2a on 85 genes, I was wondering if I need to apply the Bonferroni correction on the P-values I got for each gene?

If yes, will this work as following?

Alpha= 0.05

Genes = 85

Threshold = 0.05/85 = 0.000588

Is Significant level = P_value of each gene < 0.000588 ???

Thanks for help.

Regards,

hvk

[Ziheng]

Whether to apply a multiple test correction is not that clear in this context.

What you described is sensible, and the null hypothesis you will be testing will be that there is no gene under positive selection. Often that is not an interesting hypothesis, since it is safe to assume that some genes are under positive selection, and there is no point of testing.

To answer the question of which genes are under positive selection, I think it is probably more correct not to apply a multiple test correction.

Ziheng

[Richard Broughton]

Hi All, in a related question, we have done branch-site tests on 20 different branches of a tree with the same sequence alignment.  12 of these yield significant results uncorrected.  But only the 7 most extreme LRT values are significant after sequential Bonferroni.  If it would not be appropriate in this case to apply a correction, I am not sure how to convince a reviewer who has demanded it.  Any suggestions would be greatly appreciated.
Thanks!
Rich Broughton    

[cajawe]

Even though you've conducted just a single round of model fitting--alternative vs null for each branch--you're really asking two questions:

(1) Did positive selection occur on any of the phylogeny's 20 branches?

(2) If yes, then which ones?

For the first question, it seems wise to use a correction--20 branches means 20 tests means a solid chance of answering 'yes' to this question even if no positive selection occurred.  

For the second question, I think you can justify not using the correction so long as you recognize that (1) it's possible that some of your 12 positively selected branches represent false positives, and (2) your strength of evidence varies branch by branch.

Following Ziheng's advice to the previous message, I suppose you could argue that the first question is irrelevant——maybe you have evidence from external sources or alternative tests that positive selection definitely occurred at some undefined point in the tree, and now you're simply figuring out the where and when of it. 

[Richard Broughton]

Hi, many thanks for your helpful response.  As to question 1, sites model tests M0-M3 and M7-M8 were highly significant suggesting that selection did occur on the phylogeny.  So with the branch-site models we only interested in addressing question 2.
Thanks,
Rich