Multiple test correction in Hyphy Analysis ?

Question

[lokraj2003]

I did selection pressure analysis on viral genomes using MEME, SLAC and FUBAR on Datamonkey server. I am using p value < 0.05 for MEME and SLAC. Do I have to do any type of multiple test correction on the p-value given by MEME and SLAC ? How can I do Bonferroni Test if I have 85 sequences each with 1250 animo acids ?

Answer 1

[Brice Sarver]

Correcting for multiple tests is done based on the number of tests performed, not the size of the dataset. For example, if you split your viral genomes up into N coding loci alignments to test for selection, you'll need to correct assuming N tests.

One way to think about a Bonferroni correction is simply dividing the p-value threshold by the number of tests. You can also easily adjust a vector of p-values under different approaches using p.adjust() in R.

[lokraj2003]

Thanks Brice. I did not split the genome. So, 1250 amino acids was one gene. So, I don't need to adjust my p -value ? Just to confirm I understood your point correctly.

[Brice Sarver]

If you're taking the same dataset and testing three different, independent hypotheses (in your case, looking for different flavors of selection), you should still correct by the number of tests. It shouldn't have much of an impact with such a small number, but the same logic (accounting for type I errors that will show up by chance) will apply. Depending on the statistics, if you're looking for signatures of selection per site as opposed to per alignment, the p-value(s) reported may be corrected already - something to check. This review may help you select when/if to correct your p-values when testing multiple hypotheses, depending on what you're looking for.