Modeling and Validation of Gene Networks in Breast and Pancreatic Cancer

Omics data have been growing at an extraordinary pace and are on track to hit exabytes of data within the decade. With the ability to quickly sequence whole genomes within a day, there are many opportunities to use this data in the study of disease pathophysiology and in therapy development. With 1.9 million new cases of cancer and an estimated 610,000 deaths each year in the United States, it would be beneficial to have an integrated bioinformatic pipeline to quickly and efficiently integrate this growing amount of publicly available data in cancer related omics research. In our studies, we utilized a multi-omics approach to develop an integrated bioinformatic pipeline which combines machine learning methods, the latest bioinformatic tools, and various large omics data sets for uncovering disease mechanisms. We then took advantage of large patient cohort databases to establish a methodology for testing and validating the clinical relevance of these novel mechanisms. Using this approach, we conducted three studies in breast and pancreatic cancer. First, we focused on the oncogenic mechanisms of alcohol in the development and progression of breast cancer. We conducted a secondary analysis of our previous published transcriptome data and used our integrated pipeline to discover alcohol-regulated metabolic genes associated with oncogenic calcium signaling. The role calcium signaling in cancer promoting actions of alcohol was validated experimentally and the clinical relevance of these genes was established by their expression profiles in publicly available patient data. Our second study used a similar pipeline to find gene networks which play an important role in mediating the cytotoxic effects of liver x receptor (LXR) β activation in pancreatic cancers. This study revealed that LXRβ activity up-regulated the genes involved in pro-apoptotic fatty acid production and endoplasmic reticulum stress genes and down-regulated cell cycle genes. Many of these genes showed clinically relevant expression profiles in patient samples. Finally, our third study integrated machine learning with our established pipeline to predict targetable mechanisms in novel predicted pancreatic cancer molecular subtypes. We uncovered three novel subtypes with targetable immune associated mechanisms and subtype-specific gene expression profiles with therapeutic and prognostic implications.