@unpublished{pittir34988,
           month = {July},
           title = {Modeling and disrupting protein interactions},
          author = {Nicolas Pabon},
            year = {2018},
        keywords = {Induced fit, genomics, molecular docking, drug discovery, systems biology},
             url = {http://d-scholarship-dev.library.pitt.edu/34988/},
        abstract = {Rational drug design requires a deep understanding of protein interactions, both in terms of the structural mechanisms that regulate binding of individual molecules and the broader, pathway- and cell-level effects of disrupting protein interaction networks. This thesis presents work that stems from these ideas, discussing contributions to a number of current challenges in the field of drug discovery. First, we describe how structural flexibility is leveraged by ?selectively promiscuous? protein interfaces ? enabling them to bind specifically to several distinctly shaped ligands. Taking PD-1 as a case study, we demonstrate using molecular dynamics simulations how the flexible receptor interface recognizes conserved ?trigger? motifs on its cognate ligands? interfaces. Trigger interactions, which we show are also exploited by a recent blockbuster PD-1 inhibitor, drive the initial steps of an induced-fit binding pathway that then ?splits? into distinct, ligand-specific bound states. Second, we present a hybrid genomic and structural pipeline for genome-scale identification of protein targets for bioactive compounds. We train a random forest classifier to predict compound-target interactions from compound treatment and gene knockdown gene expression signatures in multiple cell types. Refining genomic predictions with a structure-based screen, we achieve top-10/top-100 target prediction accuracies of 26\%/41\%, respectively, on a validation set of 152 FDA-approved drugs, and validate previously unknown small molecule modulators of HRAS, KRAS, CHIP, and PDK1. Third, we present a strategy that combines transcriptomic and structural analyses with single-cell microscopy to predict small molecule inhibitors of TNF-induced NF-kB signaling and elucidate the network response. Validating two novel pathway inhibitors that disrupt the protein network upstream of IKK and NF-kB, our findings suggest that a network-centric drug discovery approach is a promising strategy to evaluate the impact of pharmacologic intervention in signaling. Last, we introduce DrugQuery (DQ), a structure-based public web server for small molecule target prediction. DQ docks user-submitted small molecules against a target library of 7957 predicted binding sites on 1245 human proteins. The server achieved a top-decile target prediction accuracy of 68\% on a validation set of 95 FDA-approved drugs and 86\% on a validation set of 102 FXR-binding compounds from the 2017 D3R Grand Challenge 2.}
}