PROJECT 3: USING NETWORKS TO SEED HIERARCHICAL WHOLECELL MODELS OF CANCER SUMMARY Knowledge of cell biology is often modeled in the form of molecular networks (aka interaction maps), consisting of sets of genegene or proteinprotein pairwise interactions. In reality, however, biological systems are not simply one large protein network, but consist of a deep and dynamic hierarchy of biological subsystems ranging across biological scales. Here, we move beyond basic interaction maps of cancer to instead use the molecular interaction data to directly infer hierarchical structure/function models of the cancer cell. These plans are enabled by a computational framework called NetworkExtracted Ontologies (NeXO), which we have recently shown is able to capture and substantially extend the known hierarchy of cellular components and processes recorded by pathway databases such as the Gene Ontology (GO). First (Aim 1), we will analyze the growing data on molecular networks to infer a Cancer Gene Ontology, representing a comprehensive, hierarchical description of the molecular complexes and pathways important for cancer. This hierarchical structure will be developed using the proteinprotein interaction data generated in Project 1, backstopped by public network data; it will provide an objective definition of a cancer cell by systematically identifying its protein modules and their interrelationships. We will next use this descriptive hierarchy to seed predictive wholecell models of cancer (Aim 2). Using the tools of deep neural networks, genetic logic will be embedded onto each complex/pathway in the cell hierarchical structure to model how perturbations to this structure give rise to cancer phenotypes. The neural network structure will be set exactly to that of the Cancer Gene Ontology assembled in Aim 1; we will then train this neural network to translate perturbations by gene mutations and drugs into predictions of cancer cell viability, as will be systematically measured in Project 2. Finally, this hierarchical model will be validated and revised by applying it to predict therapeutic responses in patientderived xenografts of head and neck and breast tumors as well as in human breast tumors from the ISPY 2 trial (Aim 3). Through execution of these aims, we will endeavor to substantially advance our knowledge of the structural and functional hierarchy of molecular pathways that underlie cancer. This hierarchy will be not only descriptive but also predictive, connecting basic knowledge of cancer pathways to a framework for using this knowledge in precision medicine.