Seminar, Hosted by Kempner Institute, Harvard University
Combining statistical genetics with machine learning towards mechanistic and therapeutic hypotheses of human disease
Harvard University
- Genetics, Bio-Informatics and
Abstract
The last two decades of human genetics have identified thousands of genetic changes that are reproducibly associated with increased risk for disease, but identifying the biological mechanisms underlying these associations remains a major challenge. This talk presents two approaches to address that bottleneck. First, a robust statistical method tests for a directional association between the activity of biologically coherent sets of proteins and disease, recovering known mechanisms and nominating opportunities for drug repurposing; the talk outlines how to combine the method with agentic optimization to accelerate discovery. Second, machine-learning predictors of the effects of genetic changes on protein function are integrated with statistical genetics methods to reveal fine-grained mechanisms of how protein-altering mutations cause disease. The talk closes with future opportunities in developing deep generative models of biological sequences and structures to advance human health.