Entering edit mode
2.7 years ago
seta
★
1.9k
Dear all,
Considering the calculation of polygenic risk score (PRS) based on several SNPs located at the same gene (without LD pruning); could you please let me know if I should do the multiple testing correction using Bonferroni when I analyzed the association of the PRS with a given trait?
Many thanks
No, you don't need to perform any sort of multiple test correction. This one is a "must have" in GWAS and does not make much sense for calculation of the PRS.
Thank you.
I agree with you as the accumulative effect of all SNPs was considered in the PRS. However, a reviewer requested doing the multiple testing correction for the association analysis of PRS with a given trait. Could you please share with me any references, mentioning that doing the multiple test correction is not required, here? I couldn't find the appropriate references.
Thank you in advance
Usually PRS performance is checked on an independent cohort. If it works on an unseen data - it is associated with the outcome.
I have never seen anyone who "validates" his/hers PRS using p-values. And it is rather difficult to perform multiple test correction if one did not perform any hypothesis tests and instead showed a predictive performance.
So I doubt there is any literature on this, it is quite unclear how to perform FDR correction in general for something which is not a set of p-values.
I think the reviewer suggested the multiple testing correction of the p-value as he/she believed several SNP of the same gene (there is the correlation among SNPs) was used for calculating the PRS. I require some reference paper to convince the reviewer that such correction is not needed. Any suggestions/comments would be highly appreciated.
Correlations between SNPs are not corrected via multiple testing correction, but by LD prunning or clumping.
Plink: Understanding LD Clumping vs Pruning