LD-pruning before logistic regression?

Question

[cs4379]

When using plink to perform a case-control logistic regression, should LD-pruning be performed beforehand? One of the assumptions with the logistic regression model is that the variables being tested are independent, and it seems like variants that are in LD would violate this assumption. If LD-pruning should be performed, what would be a good r^2 cut-off? If it shouldn't be performed, why not?

Answer 1

[chrchang523]

plink's linear/logistic regressions test one variant at a time; you only need to ensure that the covariates are independent. (After you've generated the report, --clump may be useful for organizing it into LD-based blocks, though.)