My main research focus is to develop machine learning methods that can help to decipher human disease heterogeneity. This involves combining data from multiple heterogeneous sources while addressing missing data and noise, simultaneous subtyping and feature selection in very sparse settings and more. Our contributions to machine learning include novel graph-based unsupervised feature selection methods and graphical models for subtyping in GWAS. We collaborate with clinicians to ensure that our work is relevant in the clinic.