1. Project Summary Drugs are typically developed to modulate the function of specific proteins, which are directly associated with particular disease states. Nonetheless, recent studies suggest that protein-drug interactions are promiscuous and the majority of pharmaceuticals exhibit activity against multiple, often unrelated proteins. The lack of selectivity often leads to undesired drug side effects; yet, these polypharmacological attributes can be used to develop drugs that act on multiple targets of a unique disease pathway, as well as to identify new targets for existing drugs, known as drug repositioning. Although predicting interactomes is becoming increasingly important in drug discovery, a large number of interacting molecules and highly complicated interaction patterns present significant challenges. Clearly, novel computational approaches are desperately needed to rigorously explore drug cross-reactivity. The overall goal of the proposed research is, therefore, to combine a broad scope and promises of computational systems biology, atomic-level modeling of medically relevant biomolecules and interactions among them, and heterogeneous computing using massively parallel accelerators to study drug-oriented interactomes. This innovative project comprises several components. First is to design a fully automated platform for structure-based ligand virtual screening featuring an information theory-based compound selection. By using the Maximum Entropy Method, we will be able to enhance the specificity of scoring functions for ligand ranking. Second, we plan to improve the across-proteome identification of chemically similar drug binding pockets by combining local binding site alignment with molecular docking. The advantage of this new strategy is the capability to explore a much larger space of putative cross-interactions between proteins and small organic compounds. Third, we are going to use new modeling techniques described above to reconstruct and investigate protein-drug interaction networks in the human proteome. By developing novel multi-target antibiotics, we will demonstrate that the proposed network analysis greatly expands the current opportunity space for polypharmacology and rational drug repositioning. Fourth, the scale of the task at hand as well as the level of details put an unprecedented demand for computing resources. Consequently, there is an urgent need to take advantage of modern computer architectures currently available as well as exascale supercomputers that are expected to come into production in the near future. On that account, we plan to develop high-performance codes to fully utilize heterogeneous machines equipped with massively parallel hardware accelerators, NVIDIA GPU and Intel Xeon Phi. Close collaborations with experimental and computer science groups will be part of the proposed research to make advances in this highly specialized field. The expected overall impact of this innovative proposal is that it wi...