ML-based methods for multivariate genetic association analysis

I devise methods to analyze genetic association signals and better characterize the genetic architecture of complex traits. I developed ML-MAGES, which uses neural-network-based supervised learning to account for LD-induced inflation in single-trait, SNP-level genetic associations, uses an infinite mixture model to categorize multi-trait association patterns, and aggregates and visualize association signals at gene-level. This approach substantially improves computational efficiency with large LD matrices and nonlinear inflation, achieves effect-size shrinkage comparable to or better than existing methods, and enables data-driven inference of an arbitrary number of association patterns in high-dimensional multi-trait settings.