A major challenge in oncology drug development is to translate results from large in vitro screens performed in cancer cell lines into successful drugs for patients. The first step towards this goal is to handle in vitro drug response data with rigor of genomics data and quantify them to best capture the response phenotype. With that in place, one can model in vivo response using first-principle models and obtain meaningful in vivo predictions.

In cancer, the majority of variants are unspecified, even in the most common oncogenes. I will present recent work on systematically identifying variants that drive specific molecular and clinical phenotypes. Using multi-omics datasets and functional screens in combination with structural genomics, we find that functionally-relevant clusters of mutations emerge in unique and non-obvious ways, which can pave the way to matching treatments to genomes and increasing confidence in the clinical implications of mutations.

Ovarian cancer metastasis is a complex and deadly process that involves interactions between numerous cell types in diverse microenvironments. Using bioengineered models, we are analyzing this process in vitro, with the goal of identifying mechanisms that support tumor detachment, passage in the peritoneal fluid, reattachment to new sites, and remodeling of the microenvironment. A key tool supporting our efforts is the use of data-driven modeling techniques such as partial least squares regression.

A central limitation in the development of effective therapies is our inability to predict drug-drug interactions. Using large-scale screening of drug combinations, in concert with statistical modeling, we find ‘rules’ that aid in the prediction of non-additive drug-drug interactions