The project aims at cancer biomarker discovery via searching across public datasets available worldwide – for both initial exploration and validation. The analyses should identify candidate biomarkers informative on treatment response in both public and novel cohort datasets. In case of failure to explain the differential response with individual gene profiles, the search would be complemented with a pathway-level biomarker discovery using network enrichment analysis, so that disparate gene patterns can be integrated into pathway scores. This deep learning method, recently developed by us, works by mapping patient-specific sets of mutated, differentially expressed (or methylated) genes to characteristic pathways in the global interaction network. The pathway-level markers should then contribute to creation of smaller diagnostic panels. Next, we will suggest the candidates for validation either individually or upon inclusion into biomarker panels/signatures. Our results so far indicate that this “exploration-integration-validation” framework can produce statistically and biologically valid predictors of treatment outcome.