Before or after LD clumping for PRS calculation in GWAS

Question

[K Lee]

Hello I'm new to GWAS analysis and I have a one question about LD clumping step during PRS calculation with plink. If I understand correctyly, PRS score calculation steps can be summarized as shown below.

1. association test (logistic regression for the binary phenotype)
2. LD clumping , p-value threholding (avoiding overfitting)
3. Calculating PRS by using SNPs from step 2 and betavalue from step 1

My question is, Is it wrong that LD clumping precede the association test? I think the SNPs in LD block might affect the beta values during association test quite strongly, so that i think the LD clumping should be conducted prior to the association test. Please let me know if there is any point that i misunderstood . Thank you

KJ

Answer 1

[Sam]

1. Association testing are done on each SNP independently. So it would not affect the beta
2. If you do prunning before association (you cannot do clumping without p-value), you will not capture the most associated SNPs with your trait of interest You can refer to the PGS tutorial paper: https://www.nature.com/articles/s41596-020-0353-1

[K Lee]

1. I think i misunderstood logistic regression of association test. Now I understand.
2. you're right. Thank you for your helpful reply sam!