Uncovering Molecular Subtypes of Heart Disease with AI

The article describes how the author built an AI portal to prove that heart disease is not a one-size-fits-all condition. By integrating multi-omics data, the model identified three distinct molecular subtypes of heart disease with unique biomarker patterns and treatment targets.

Details

The author was motivated to build an AI portal to prove that heart disease is not a monolithic condition, as two patients with the same diagnosis can respond very differently to the same treatment. Coming from a microbiology background, the author was fascinated by the invisible cellular mechanics driving heart disease. Instead of using a single dataset, the author decided to integrate four biological 'chapters' - genomics, transcriptomics, proteomics, and metabolomics - to get a more comprehensive view. By testing each integration layer sequentially using MOFA+, the author found that the clustering quality improved by 178% when all four data layers were combined, proving that the biological signals are synergistic. The model identified three distinct molecular subtypes of heart disease: Energy Metabolism (genetic predisposition and mitochondrial dysfunction), Inflammatory (autoimmune-mediated cardiac inflammation), and Fibrotic (pathological cardiac fibrosis). Each subtype has unique biomarker profiles and potential treatment targets, highlighting the need for a more personalized approach to heart disease management.