Saddlepoint approximation
When analysing binary traits, we recommend running single-SNP analysis using a saddlepoint approximation (SPA). Classical logistic regression may result in inflation of type 1 error when testing phenotypes with case:control imbalance. The SPA implemented in LDAK guards against false positives and can robustly test SNPs for association, even when testing data with highly imbalanced case:control ratio.
The SPA is included in LDAK by default when running --logistic
, for example:
./ldak.out --logistic out --bfile data ----pheno phenofile --covar covfile
The SPA can be turned off by specifying --spa-test NO
, in this case LDAK switches to classical logistic regression.