Improving Modeling by Learning from Details of High Accuracy Protein Structures

通过学习高精度蛋白质结构的细节来改进建模

• 批准号: 8547080
• 负责人: Paul Andrew KARPLUS
• 金额: $ 20.07万
• 依托单位: OREGON STATE UNIVERSITY
• 依托单位国家: 美国
• 财政年份: 2008
• 资助国家: 美国
• 起止时间: 2008-08-01 至 2016-07-31
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/8547080
• 关键词: Accounting; Automobile Driving; Behavior; Benchmarking; Biomedical Research; Catalysis; Cells; Computers; Crystallography; Databases; Development; Drug Formulations; Environment; Enzymes; Experimental Models; Funding; Generations; Glean; Goals; Homologous Gene; Investments; Learning; Libraries; Life; Machine Learning; Maintenance; Methodology; Methods; Mining; Modeling; Molecular Conformation; Peptides; Pharmaceutical Preparations; Process; Protein Conformation; Protein Structure Initiative; Proteins; Research Personnel; Resolution; Resources; Risk; Side; Source; Structure; Technology; Testing; Time; Torsion; United States National Institutes of Health; Validation; Variant; Vertebral column; Work; base; cost; design; disease-causing mutation; falls; improved; inhibitor/antagonist; innovation; knowledge base; molecular mechanics; next generation; predictive modeling; protein function; protein structure; protein structure prediction; tool; trend; ultra high resolution

DESCRIPTION (provided by applicant): The functions of proteins depend exquisitely on their structure, with details at the 0.1 ¿ scale influencing enzyme catalysis, disease-causing mutations, and drug recognition. For this reason, having detailed and accurate structures of proteins is a cornerstone of modern biomedical research, and the NIH funded the Protein Structure Initiative with the goal of obtaining models for every protein structure with an accuracy approaching that of a high-resolution crystal structure. Current technology for template-based modeling is powerful, but cannot yet deliver "near-crystal-structure" quality. Tests show that the best minimization routines still fall short of consistently producing protein models for close homologs that approach within ~1 ¿ rmsd of the 'native' structure as ultimately revealed by crystal structures. To help break through this 1 ¿ barrier, during the previous period of support we used ultrahigh-resolution structures to create a library of conformation- dependent ideal geometry functions for the protein backbone, and showed that its use improves the quality of protein crystal structures and holds promise to improve template-based model refinement. We also discovered that ultrahigh-resolution crystal structures are a rich source of details about protein structure that are not accurately attainable from structures in the ~1.5-2 ¿ resolution range and thus have not yet been fully accounted for in current energy functions. Here, our central hypothesis is that a major step forward in template-based modeling accuracy will come from identifying and explicitly taking into account detailed features of protein covalent geometry, conformation and non-covalent packing interactions that have not yet been characterized, and can now be gleaned from the study of highly accurate ultrahigh-resolution protein structures. The overall goal of our proposal is to mine such information so it can be used to improve the accuracy of predictive modeling. With many ultrahigh-resolution structures now available, the time is ripe to achieve this goal by pursuing three specific aims related to (1) extending the impact of the 'ideal geometry function' paradigm by creating, optimizing, and implementing conformation- dependent libraries accounting for peptide planarity, side chains, and cis-peptides, (2) mining ultrahigh- resolution crystal structures to glean information for next-generation empirical energy functions, and (3) analyzing ultrahigh-resolution protein structures solved in varying environments to produce a set of benchmark test cases and developing residue level assessment tools to use with these test cases to evaluate and hone template-based modeling refinement applications. This proposed work is low cost and low risk, and has a high likelihood of substantial impact as it provides basic information that can be widely incorporated into predictive and experimental modeling applications to improve their accuracy. It is also distinct from major efforts being invested into template-based modeling. Introducing this greater level of realism is a prerequisite to improving the refinement step of template-based modeling and achieving the goals of the Protein Structure Initiative.

描述（由申请人提供）：蛋白质的功能完全取决于其结构，0.1 级的细节会影响酶催化、致病突变和药物识别。因此，拥有详细而准确的蛋白质结构是现代生物医学研究的基石，NIH 资助了蛋白质结构计划，目标是获得每种蛋白质结构的准确模型 接近高分辨率晶体结构。当前基于模板的建模技术很强大，但尚无法提供“接近晶体结构”的质量。测试表明，最好的最小化程序仍然无法持续生成密切同系物的蛋白质模型，这些模型接近晶体结构最终揭示的“天然”结构的约 1 ¿ rmsd 范围内。为了帮助突破这一 1 ¡ 障碍，在之前的支持期间，我们使用超高分辨率结构为蛋白质主链创建了构象依赖的理想几何函数库，并表明它的使用提高了蛋白质晶体结构的质量，并有望改善基于模板的模型细化。我们还发现，超高分辨率晶体结构是有关蛋白质结构细节的丰富来源，这些细节无法从〜1.5-2 ¿分辨率范围内的结构中准确获得，因此尚未在当前的能量函数中得到充分解释。在这里，我们的中心假设是，基于模板的建模准确性的重大进步将来自于识别并明确考虑蛋白质共价几何的详细特征， 尚未表征的构象和非共价堆积相互作用，现在可以从高精度超高分辨率蛋白质结构的研究中收集。我们提案的总体目标是挖掘此类信息，以便将其用于提高预测建模的准确性。现在有许多超高分辨率结构可用，通过追求以下三个具体目标来实现这一目标的时机已经成熟：(1)通过创建、优化和实施考虑肽平面性、侧链和顺式肽的构象依赖库来扩展“理想几何函数”范式的影响，(2)挖掘超高分辨率晶体结构以收集下一代经验能量的信息 (3) 分析在不同环境中解决的超高分辨率蛋白质结构，以生成一组基准测试用例，并开发残留水平评估工具，与这些测试用例一起使用，以评估和磨练基于模板的建模细化应用。这项拟议的工作成本低、风险低，并且很有可能产生重大影响，因为它提供了可以广泛纳入预测和实验建模应用程序以提高其准确性的基本信息。它也不同于基于模板的建模所投入的主要精力。引入这种更高水平的真实感是改进基于模板的建模的细化步骤和实现蛋白质结构计划目标的先决条件。