p.adjust with n < than number of tests

Question

[yogacacarolom]

I would like to apply the p.adjust function in R where n is < number of p-values. The real number of independent tests is lower than the number of p-values as it cames from genomic data with Linkage Desequilibrium:

However, the p.adjust function do not allows it: number of comparisons, must be at least length(p).

Someone knows how to change this default in function or other generic function to accomplish similar work? Thank you!

Followed Steps:

1 - 3242 tested markers = 3242 p-values

2 - Inferred Meff is: 1096 (http://simplem.sourceforge.net/ procedure)

Now I need to estimate the corrected treshould or corrected p-values based on Meff.

I am not sure which multiple test correction strategy fits better or how to apply it in my data.

[Devon Ryan]

I don't think the methods implemented in p.adjust() are appropriate for your use case.

[yogacacarolom]

I described more details about my analysis. Some idea of suitable strategy from now?

[Devon Ryan]

Don't they propose in their 2010 paper that one can use Meff in bonferroni or Sidak correction? Those are simple enough to just directly calculate (they'd be a single line in R).

[yogacacarolom]

Yes: "The idea of Meff-based approaches is simple: "filter out" the correlation among tests, leaving only the effective number of independent ones, Meff, and then use the Bonferroni or the Šidák correction by replacing M with Meff in the corresponding formula." (GAO et. al, 2010) -> http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2796708/#R4. Would you think that it can be applied to "FDR" also?

[Devon Ryan]

Both Sidak and Bonferroni thresholds will be similar (~4.68e-5 and ~4.56e-5, respectively, assuming the stereotypical 0.05 threshold), though the former is probably a bit more preferable to use simply due to it not being as overly conservative.