Data-driven Computational Modeling and Refinement of Protein Structures on Genomic Scales

项目摘要/摘要： 在我们对分子水平的生物系统的理解中，一个关键的剩余差距是如何从结构上 诠释“暗”蛋白质家族--实验结构未解决的蛋白质家族部分 确定技术和同源建模不可用。近四分之一的蛋白质家族是 目前是黑暗的，分子构象完全未知，这个缺口可能会进一步扩大 随着没有注释结构的新蛋白质序列的快速积累。关键的挑战是现在 如何弥合这一差距，以全面了解生物学和疾病，从而为 到基因组水平的基于结构的药物设计。计算蛋白质模型在这项工作中起着关键作用 由于它的可扩展性和全基因组的适用性。我的实验室专注于这一领域的开发和应用 新的数据驱动的计算建模和改进方法，以提高准确性和覆盖率 基因组水平上的蛋白质结构预测，与同源性无关。未来的研究重点是改进 基于深度学习的残基间多尺度从头建模的无同源蛋白质折叠 交互，通过制定新的远程算法来增强低同源性线程或折叠识别 尽管进化关联度较低的模板识别以及用于高分辨率的开发方法 广义集成搜索引导的约束结构精化驱动计算模型 接近实验精度。将进行蛋白质组范围的计算建模和改进工作， 利用我们对大规模超级计算基础设施的独特访问，构建高置信度模型 涵盖黑色蛋白质家族，将被组织在一个数据库中供公众查阅。这一全面的 结构注释数据库将阐明黑暗的结构、功能和相互作用 蛋白质组，在药物发现和人类健康方面具有广泛的影响。软件和Web服务器将免费 传播以帮助世界各地的生物医学研究人员将这些方法应用于他们的特定 研究问题，因此计算建模对生物学和基础研究的影响成倍增加 医药。我的研究计划将涉及与NIGMS支持的其他调查人员的密切合作， 为下一代研究人员创造培训机会，包括来自代表性不足的成员 并促进结构生物信息学和计算生物学的未来研究进展。

PROJECT SUMMARY/ABSTRACT: A key remaining gap in our understanding of biological systems at the molecular level is how to structurally annotate the “dark” protein families—the portion of protein families unsolved by experimental structure determination techniques and inaccessible to homology modeling. Nearly a quarter of protein families are currently dark, where molecular conformation is completely unknown and this gap is likely to expand further with the rapid accumulation of new protein sequences without annotated structures. The key challenge is now how to bridge this gap to gain a comprehensive understanding of biology and disease, thereby paving the way to structure-based drug design at genomic scale. Computational protein modeling plays a key role in this effort due to its scalability and genome-wide applicability. My laboratory focuses on the development and application of novel data-driven computational modeling and refinement methods to increase accuracy and coverage of protein structure prediction on genomic scale irrespective of homology. Future research focuses on improving homology-free protein folding using multiscale de novo modeling driven by deep learning-based inter-residue interactions, enhancing low-homology threading or fold recognition by formulating new algorithms for remote template identification despite low evolutionary relatedness, and developing methods for high-resolution restrained structure refinement guided by generalized ensemble search for driving computational models to near-experimental accuracy. Proteome-wide computational modeling and refinement effort will be conducted, leveraging our unique access to large-scale supercomputing infrastructure, to build high-confidence models covering the dark protein families, which will be organized in a database for public access. This comprehensive database of structural annotations will shed light on the structures, functions, and interactions of the dark proteome, with broad implications in drug discovery and human health. Software and web servers will be freely disseminated to help worldwide community of biomedical researchers to apply these methods to their specific research problems, thus multiplying the impact of computational modeling on basic research in biology and medicine. My research program will involve close collaborations with other NIGMS-supported investigators, create training opportunities for the next generation of researchers including members from underrepresented groups, and foster future research advances in structural bioinformatics and computational biology.