Project 4: Summary While genomewide association studies (GWAS) have linked many genetic loci to complex diseases, the loci mapped thus far account for a small fraction of the total genetic variation affecting these phenotypes. This limitation is common to both human GWAS and GWAS in model organisms such as the heterogeneous stock (HS) rats that are the focus of this center. To better capture the genetic signal, we (laboratory of Project 4 Director Trey Ideker) and many others have argued that GWAS results must be integrated with fundamental knowledge of molecular and cell biology, as captured by biological network models. To this end, we will create computational analysis tools to synthesize GWAS data with molecular network information, advancing the current state of computational genetic analysis. These tools will be benchmarked and applied in the context of diverse drug abuserelated behavioral phenotypes studied by Projects 1, 2, and 3, as well as phenotypes being studied by the separately funded “affiliated grants.” Work will progress along three Specific Aims: First, we will mature and apply the technique of network propagation for gene association analysis. In recent studies, network propagation has been shown to substantially boost power to identify reproducible and functional genetic associations, while also providing concrete hypotheses as to the underlying molecular mechanisms transmitting genotype to phenotype. We will also extend this method to integrate Transcript Wide Association Study (TWAS) approaches. Second, we will develop molecular networks as a tool for translation of GWAS results between rat and human studies related to drug abuse. This aim will rely on the conservation of molecular pathways between species to find overlapping mechanisms associated with both rat and human phenotypes. Third, we will build on the above results to develop a hierarchical reference model of pathways in which genetic variation is associated with drug abuse. We will explore the extent to which this pathway hierarchy can be used to structure a deep artificial neural network (ANN) for translation of genotype to phenotype. This system, based on a previously published prototype in budding yeast, will be extended along significant lines for application to mammalian genetics. If successful, it will not only make accurate predictions of phenotype from genotype, it will also be interpretable and fuel mechanistic hypotheses relevant to the development of novel treatments for drug abuse.