All Times ET
Keywords: driver genes, driver proteins, cascading proteins, pan-platform integration, spike-and-slab prior, hierarchical models
Rapid technological advances have made various types of genomic, epigenomic, transcriptomic and proteomic data available, each offering a different, partly independent, and complementary, high-resolution view of the genome. Modeling and inference using such data is challenging, due to both high-dimensionality and structured dependencies, but can be an effective tool for understanding the complex biological processes characterizing a disease. We propose fiBAG, an integrative hierarchical Bayesian framework for modeling the fundamental biological relationships underlying such cross-platform molecular features. Using Gaussian process (GP) models, fiBAG identifies mechanistic evidence for covariates based on their corresponding upstream information. By mapping said evidence to prior inclusion probabilities, a calibrated Bayesian variable selection (cBVS) model is built to identify genes/proteins associated with the outcome of interest. Using synthetic datasets, we illustrate how integrative methods like cBVS have higher power to detect disease related markers than non-integrative approaches. Further, we perform a pan-cancer analysis of 14 different cancer datasets from the Cancer Genome Atlas (TCGA) to identify genes and proteins associated with cancer stemness and patient survival and their potential cellular mechanisms. Our findings include several known associations, such as the RPS6KA1 gene/p90RSK kinases in gynecological cancers, alongside some interesting novelties like EGFR in gastrointestinal cancers. To ensure accessibility for users belonging to all relevant knowledge domains, we offer a Shiny dashboard summarizing all these results and a dynamic repository containing reproducible materials.