Molecular modeling and machine learning for protein structures and interactions

点击翻译按钮获取中文摘要

PROJECT SUMMARY / ABSTRACT Structural biology provides a powerful lens through which to view living systems. With advances in algorithms and computing, molecular simulations have begun to complement traditional experimental approaches as tools for discovery. At the same time, data-intensive machine learning approaches are becoming increasingly important in biology, fueled by the rapid growth in high-throughput experimentation. Research in my laboratory applies techniques from structural biology, molecular simulation, and machine learning to design new protein structures and predict protein interactions. We design new protein structures in order to better understand the principles of protein folding and to create highly stable and robust molecular scaffolds for a range of biomedical applications including multivalent display of binding or signaling domains, hosting of binding or catalytic sites, and use as building blocks to assemble higher-order complexes. We predict protein interactions in order to better understand the principles of macromolecular recognition and to gain insight into the process by which the adaptive immune system discriminates self from non-self in the context of infectious and autoimmune diseases and cancer. Our research during the project period will be directed toward two broad goals: de novo design and functionalization of tandem repeat proteins, and prediction of peptide-MHC recognition by T cell receptors (TCRs). The proposed protein design work builds on our recent progress designing circular tandem repeat proteins with a range of repeat numbers and diameters and applying these designs as multivalent display scaffolds for the presentation of binding and signalling domains. Our TCR studies leverage the tools we have recently developed to model—structurally and bioinformatically—repertoires of T cell receptors and their peptide:MHC specificity. Looking ahead, I am optimistic that by combining atomically-detailed molecular simulations and data-intensive machine learning techniques we will be able to generate designed protein constructs and predictive algorithms that have a significant positive impact on human health.