ARC23 Speaker Jishnu Das

Using machine learning and network approaches to elucidate molecular phenotypes underlying Covid-19 outcomes: Integration of a structurally resolved protein network with signaling cascades uncovers immunomodulatory molecular phenotypes in infectious disease

Jishnu Das, assistant professor, Pitt Departments of Immunology and Computational & Systems Biology
Director, Computational Immunogenomics Core, Center for Systems Immunology, University of Pittsburgh School of Medicine

Abstract
Past work by us and others has demonstrated the critical importance of incorporating corresponding structural information in the integration of Mendelian mutations with protein networks, to elucidate molecular phenotypes underlying the corresponding genetic disorders, with high sensitivity and specificity. However, while Mendelian disorders are typically monogenic and often attributable to a handful of high penetrance mutations, the role of genomic variation in infectious disease is far more nuanced, as specific pathogens cause these disorders and the genetic variants are immunomodulatory. Here, we present a novel pipeline that integrates a structurally resolved reference human protein interactome with signaling cascades inferred from expression and chromatin accessibility data to uncover cell-type-specific immunomodulatory molecular phenotypes in infectious disease. We used Covid-19 as an exemplar given the availability of deep genotype data from the Covid-HGI, and corresponding cellular and molecular data. However, the framework is broadly applicable across infectious disease contexts.

We combined genotype data for COVID-19 disease occurrence and severity with a structurally resolved reference human protein interactome network to identify coding mutations at and away from the interfaces of specific protein-protein interactions. We then compared the cell-type-specific functional impact of these variants to non-coding regulatory variants that attain/do not attain genome-wide significance. The functional impact of these classes of mutations on different specific cell types was inferred via network propagation of the effects of these mutations on cell-intrinsic and cell-extrinsic signaling networks inferred from scRNA-seq and scATAC-seq data. Mutation effects within cell types were also ensembled to evaluate the relative importance of different cell types in Covid-19 pathogenesis. We recapitulated the well-known roles of classical and non-classical monocytes in Covid-19 pathogenesis. However, surprisingly, we found that a large component of the signal was driven by only a handful of prioritized coding variants at specific protein interaction interfaces. These were functionally as important as genome-wide-significant regulatory variants, and both sets were far more important than all other classes of variants. Of particular interest, were three variants prioritized by our pipelines that provide novel mechanistic insights into how genotypic differences modulate the host immune response to Covid-19. Our framework is broadly applicable to other infectious disease contexts.

Biography
Jishnu Das completed an undergrad degree in Bioengineering at the Indian Institute of Technology Kanpur followed by graduate studies in Computational Biology at Cornell University. After a short postdoc at the Ragon Institute of MGH, MIT & Harvard/MIT, Jishnu started his lab in January 2020 as an Assistant Professor at the Center for Systems Immunology. His research involves using the use of machine learning, network systems and functional genomic approaches to integrate multi-omic datasets with biological networks to identify molecular phenotypes in infectious, autoimmune, and alloimmune diseases. His research is currently supported by 3 R01/R01-equivalent awards as PI/MPI and 9 other federal grants including NIH and DoD awards.