Biomedical data sharing and analysis at scale : privacy, compaction, and integration

Abstract

Recent advances in high-throughput experimental technologies have led to the exponential growth of biomedical datasets, including personal genomes, single-cell sequencing experiments, and molecular interaction networks. The unprecedented scale, variety, and distributed ownership of emerging biomedical datasets present key computational challenges for sharing and analyzing these data to uncover new scientific insights. This thesis introduces a range of computational methods that overcome these challenges to enable scalable sharing and analysis of massive datasets in a range of biomedical domains. First, we introduce scalable privacy-preserving analysis pipelines built upon modern cryptographic tools to enable large amounts of sensitive biomedical data to be securely pooled from multiple entities for collaborative science. Second, we introduce efficient computational techniques for analyzing emerging large-scale sequencing datasets of millions of cells that leverage a compact summary of the data to speedup various analysis tasks while maintaining the accuracy of results. Third, we introduce integrative approaches to analyzing a growing variety of molecular interaction networks from heterogeneous data sources to facilitate functional characterization of poorly-understood genes. The computational techniques we introduce for scaling essential biomedical analysis tasks to the large volume of data being generated are broadly applicable to other data science domains.