#4 · Genomics

Integrating diverse variant and annotation data to improve genetic risk prediction

The dominant thread is improving genetic risk or trait prediction by incorporating information beyond standard common-variant effects. RICE combines common-variant ensemble learning with rare-variant gene-level testing and penalized regression, evaluated across diverse populations. IFAM extends linear mixed models with multiple random effects to integrate many functional annotation types, improving genomic prediction accuracy for complex traits. The type 1 diabetes study pairs genome-wide association and fine-mapping of 20,355 cases, yielding 160 risk signals, with the machine learning model T1GRS, which improves risk prediction. The ECG foundation model is less directly genomic: it predicts eight cardiovascular diseases with a mean AUROC of 0.930 and supports genetic factor discovery. What is new is the mode of integration: rare and common variants within one framework, annotation types as separate random effects, and machine learning layered onto genetic association. The statistical frameworks differ, so the shared element is the prediction objective rather than a single method class.