Computational and Experimental Investigation and Design of Protein Interaction Specificity

蛋白质-蛋白质相互作用传递信息，塑造细胞结构，组装复合物，并使

Protein-protein interactions transmit information, shape cell structure, assemble complexes, and enable chemical transformations that support life. Mapping and decoding the human interactome to establish which interactions occur, what functions they support, and how interactions are altered in disease are critical goals for biology. There is also a biomedical imperative to learn to inhibit or modulate protein interactions for discovery research and the development of new therapies. This proposal presents an integrated program of computational and experimental studies of protein-protein interactions that involve short linear motifs (SLiMs) binding to modular, structurally conserved interaction domains. SLiM are abundant, with estimates of more than 105 binding motifs in the human proteome, and they play critical roles in signal transduction and the assembly of structural and regulatory complexes that are implicated in disease. The domains that bind to SLiMs, such as EVH1, TRAF, SH3, WW, etc., occur in many copies in the proteome due to the expansion of paralogous families by domain duplication and divergence. This research program will address two key questions. (1) The paralog specificity question: How do the interactions made by paralogous protein domains overlap vs. differ, and how are distinct binding profiles encoded in similar sequences and structures? Answering this will provide currently missing links in the interactome and support the prediction and design of paralog-specific interactions, which will improve our knowledge of disease pathways and how to target them. (2) The SLiM specificity question: What sequence/structure features determine SLiM binding and how is this regulated? Learning the features that distinguish real interactors from myriad motif-matching false positives in the proteome will uncover mechanisms of SLiM recognition and support the prediction of new interactions. This proposal focuses on developing new methods and models that will be applied to study biomedically important SLiM-binding EVH1 and Atg8-like domains. EVH1 domains are found in proteins that bind to proline- rich motifs, including members of the Ena/VASP family that regulate cancer cell invasion and metastasis. Atg8-like proteins are critical for autophagy and participate in forming the autophagosome and recruiting cargo for degradation by binding to selective autophagy receptors. Increased or decreased autophagy contributes to many diseases via poorly understood mechanisms. The proposed studies will combine high-throughput interaction mapping using experimental cell-surface display screening with data-driven modeling using deep learning to support the detection, prediction, and design of new interactions. The screening-plus-modeling approach will reveal new interaction partners for each family that broaden our understanding of cell biology, elucidate mechanisms of specificity, and provide new techniques for designing selective inhibitors of these and other protein-protein interactions.