As a preamble, I am somewhat knowledgeable in statistics, but have reached a problem in that I'm not sure how to go about statistical significance in GWAS based on the literature alone (I've read a couple papers but it's a little dense given my limited skill set and I'm not sure how the numbers really line up in regard to the equations often posted).
Essentially I have 143 cell lines and 50,291 SNPs being investigated. I have read that you might take the alpha value of 0.05 and divide by the number of SNPs to use that as a cutoff (so in this case, 0.05 / 50,291 ~ 9.94x10^-7). Unfortunately that is a bit too strict given the 143 cell lines with genetic data I have available.
The statistical testing software I am using is providing a multivariate ANCOVA, controlling for biological sex, age, race, ethnicity, and family (these being covariates), with a single response variable (R1). My "main" regression model output statistic and p-value are based on Pillai's Trace. As stated above, 9.94x10^-7 is a bit too strict, given that my "best" p-values are around 1x10^-5. But should I be considering those significant at all? Why or why not? Are there steps in this process that I am completely forgoing?
Thank you so much!
Did you try calculating adjusted p-values using B-H or B-Y? I guess you can use adjusted p-values (BY or BH).
Is B-H referring to Benjamini-Hochberg and B-Y to Benjamini-Yekutieli? I haven't tried calculating them... So would I set my false positive rate as 10% (for B-H)? That would put one of my example "best" P-values at 1x10^-5
adjusted p-rate cutoff is 0.05 (in general). Some prefer 0.01. B-H and B-Y are as you mentioned above. You may use qvalue package in R.
Okay, totally just curious here... So say my raw p-value is 1x10^-5. So then I divide that by the number of SNPs (50,291) for my corrected p-value (Bonferroni), as ~1.988x10^-10. Then I use that as my p-value in the B-H procedure?
Yes, calculate the nominal p-value at which FDR is 5%, 1%, etc., and then use that as the cut-off.
Another thread to assist in linking these up: