The Causal Pivot: A Structural Approach to Genetic Heterogeneity and Variant Discovery in Complex Diseases

Abstract

We develop a structural causal methodology to analyze genetic heterogeneity in complex diseases. Our method, the Causal Pivot (CP), exploits one known cause to detect the contribution of a second suspected cause. We employ polygenic scores (PRS) as a known cause; we consider selected rare variation (RV) and RV ensembles as candidate causes. We condition on disease to determine a method that incorporates outcome induced association. We derive a conditional maximum likelihood procedure for both binary and quantitative traits and develop a likelihood ratio test (CP-LRT). We demonstrate the CP-LRT power and robustness to misspecification of the RV rate compared to alternatives. We then use data from the UK Biobank (UKBB) to examine performance. As demonstrations, we consider hypercholesterolemia (HC, LDL-c direct >= 4.9 mmol/L; nc=24656), breast cancer (BC, female, ICD10 C50; nc=12479), and Parkinsons Disease (PD, ICD10 G20, nc=2940). For PRS we use UKBB data; for RV we select ClinVar pathogenic and likely pathogenic variants in known disease genes. For HC we focus on LDLR, for BC we focus on BRCA1, and in PD we consider GBA. Using CP-LRT we detect RV signals for all three demonstration diseases. Cross-disease analyses serve as comparators. We extend our CP-LRT approach using matching and inverse probability weighting to address potential confounding driven by ancestry. Finally, we present a CP oligogenic load analysis in PD using lysosomal storage. Taken together, our work demonstrates the versatility of the PRS-RV CP as a method of inference and discovery in complex disease genetics.