#23 · Genomics

Validity checks expose inflated false positives in association and replication analyses

The dominant thread is whether apparent associations survive explicit validity checks. In synthetic and 14 brain datasets, surrogate variable analysis inflates false positives and lowers cross-cohort reproducibility. ScreenDMT extends the directional MaxP test to multi-study replication, with proven FDR validity supported by simulations, and is applied to blood-based metabolite data across three cohorts in sleep disorders. A biobank G×E analysis shows that power and log transformation of the phenotype scale explain 45% and 23% of PGS×sex interactions, so scale choice can generate apparent interactions. The Multi-scale Copula Index is an association measure that avoids density estimation and tuning, with benchmarks extending to cancer genomics. A gradient-boosting Meniere disease classifier (test AUC 0.843) is a predictive application rather than an inferential one. What is new is framing each method around a defined validity question, such as false-positive control, replication, or phenotype-scale artefact, rather than general performance alone.