A scalable invariant causal discovery framework reveals direct genetic and clinical determinants of complex human traits

I-GCM identifies direct causal genes and clinical risk factors in UK Biobank, including JAK1 as a direct causal gene for severe COVID-19 susceptibility and severity.

• Why it matters: Precision medicine needs direct causal determinants, but association methods can’t separate cause from consequence and existing causal discovery is limited by assumptions or computational limits in high dimensions.
• What they did: The scalable Invariant Generalized Covariance Measure combines PC-simple screening with a generalized covariance measure computed via gradient-boosted trees to test invariance across environments. In simulations up to 3,000 variables and 50,000 samples, it achieved PPV ~0.84–1.00 and specificity ≥0.99.
• The result: In UK Biobank, I-GCM produced direct causal gene and clinical factor findings for COVID-19 susceptibility and severity and for HDL and triglyceride levels, with many supported by literature and pathway enrichment. It also links JAK1 to severe COVID-19 in a way consistent with baricitinib use.

AI summary of the abstract · may contain errors · not medical advice · read the paper.