Next-generation sequencing has significantly advanced our understanding of biology by providing a unique genomic perspective of the molecular state of human disease. However, combining multiple ‘omics’ measurements into biologically-relevant statistical computing frameworks to define causal molecular underpinnings of disease remains a significant challenge. To address these issues, we developed novel feature learning approaches that enhance quantitative assessment of annotated tissues from The Cancer Genome Atlas. Herein, we demonstrate the utility of collapsing molecular signals, from five different -omics platforms, into integrated metagenes that are highly informative across roughly 8,200 tumors, encompassing 22 cancer types. Our <i>a priori</i> biological-knowledge and data-driven network-based approaches improve performance and interpretability of both deep learning and probabilistic programming strategies, revealing novel network-driver genes underlying cancer type specific etiology and pan-cancer patient survival. These results demonstrate the usefulness of our methodologies in addressing currently unresolved statistical issues associated with the analysis of high-dimensional biological data.

Ensemble computational intelligence reveals novel molecular signatures of cancer biology and pan-cancer survival

256th National Meeting (2018)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.