Integrated Prediction and Validation of Protein Structures

 描述（由申请人提供）：三维蛋白质结构的知识在生物医学研究中是不可或缺的。蛋白质的结构和功能是密切相关的，因此结构有助于药物发现，有助于蛋白质-蛋白质相互作用的研究，告知诱变分析，指导蛋白质工程和新蛋白质的设计，并为理解疾病的分子基础提供基础。然而，在基因组时代可用的蛋白质序列的数量远远超过了X射线晶体学和核磁共振（NMR）光谱学的主要实验结构确定技术的能力，导致大量的序列-结构缺口。我们通过开发和传播新的蛋白质结构建模工具来解决这一不断扩大的差距。这个更新项目是计算建模（Cheng）和实验结构生物学（坦纳）专家之间的新合作。我们计划开发创新的、集成的机器学习（例如，深度学习）、数据挖掘和统计建模方法，以解决基于模板的结构建模和无模板（从头算）结构建模中的主要挑战。我们将这些工具应用于醛脱氢酶（ALDH）超家族中的酶，这是一组参与许多重要生物过程并因突变而与许多疾病有关的酶。ALDH模型将使用X射线晶体学和生化分析进行实验验证。此外，我们将结合联合收割机的建模能力，我们的结构输入输出隐马尔可夫模型与实验小角X射线散射（SAXS）预测的三级结构的大型多域蛋白质。在这个项目中，计算和实验科学的整合使我们在结构建模空间中处于独特的地位。

 DESCRIPTION (provided by applicant): Knowledge of three-dimensional protein structure is indispensable in biomedical research. Protein structure and function are intimately linked, and thus structure facilitates drug discovery, aids investigations of protein-protein interactions, informs mutagenesis analysis, guides protein engineering and the design of new proteins, and provides a foundation for understanding the molecular basis of disease. However, the number of protein sequences available in the genomic era far exceeds the capacity of the main experimental structure determination techniques of X-ray crystallography and nuclear magnetic resonance (NMR) spectroscopy, resulting in a substantial sequence- structure gap. We address this ever-widening gap by developing and disseminating novel protein structure modeling tools. This renewal project is a new collaboration between experts in computational modeling (Cheng) and experimental structural biology (Tanner). We plan to develop innovative, integrated machine learning (e.g., deep learning), data mining and statistical modeling methods to address major challenges in both template-based structure modeling and template-free (ab initio) structure modeling. We will apply these tools to enzymes in the aldehyde dehydrogenase (ALDH) superfamily, a group of enzymes that are involved in numerous important biological processes and implicated in many diseases due to mutations. The ALDH models will be experimentally validated using X-ray crystallography and biochemical assays. Furthermore, we will combine the modeling power of our structural Input-Output hidden Markov model with experimental small- angle X-ray scattering (SAXS) to predict the tertiary structures of large multi-domain proteins. The integration of computational and experimental sciences in this project positions us uniquely in structure modeling space.