Data-driven understanding of human disease: from machine learning methods to biological discoveries

Abstract:
How does the same DNA sequence lead to such different cells in the brain versus the lungs? What genomic signals encode our predisposition to Parkinson's disease? Why and how do scientists use worms to study human disease? I will discuss these questions with a focus on development and application of machine learning methods, including deep learning, Bayesian, and semi-supervised approaches, for biomedical data.

More specifically, I will address a key challenge in biomedical science -- development of a complete understanding of the genomic architecture of disease. Yet the increasingly wide availability of genomic data (including whole genome sequencing and expression) has far outpaced our ability to analyze these datasets. Challenges include interpreting the 98% of the genome that is noncoding (sometimes referred to as 'junk' DNA), detangling genomic signals regulating tissue-specific gene expression, mapping the resulting genetic circuits in disease-relevant cell types, and, finally, integrating the vast body of biological knowledge from model organisms with observations in humans. In my seminar, I will discuss methods that we have developed to address these challenges, and present their applications to autism, Parkinson's, and cardiovascular disease.