Accurate Molecular Modeling in Structural Genomics

描述(由申请人提供)：本提案的总体目标是继续提高结构基因组学同源建模的准确性。这将通过两个主要方向的进展来实现：(A)更好地对实验者发现的蛋白质结构域进行分类，以及(B)更好地改进蛋白质模型，使其更接近实际结构。随着已知结构的发现，对它们进行分类将使实验者能够更好地评估他们的进展，并将其与其他小组的工作联系起来。它还将提供用于同源建模和折叠识别的准确多结构比对的有价值的数据库。更好的改进将包括在详细的全原子水平上开发有利于天然蛋白质结构的能量函数，以及从初始模型向真实结构移动的更强大的方法。除了获得更精确的同源模型外，这两个进展都将在所有分子功能研究中具有深远的应用，包括配体结合的建模、蛋白质-蛋白质相互作用的建模以及更广泛的蛋白质功能模拟。我们认为，成功实现以下具体目标将是取得这一重大进展的最佳途径： (1)基于我们改进的结构比对方法Structal，开发了一种可靠的领域分类方法，该方法在六种不同方法的研究中表现得出人意料地好。我们将使用我们为这笔赠款提供的计算机资源每周更新数据。 (2)使用作为AIM(1)的一部分进行的常规结构搜索来产生多个结构对齐，这些对齐将基于我们用于多个结构叠加和对齐冲突解决的方法。(3)利用扭角空间和笛卡尔空间中的正交正交模生成数千个接近自然的诱饵。这些诱饵集将被生成为具有小的局部变形，但与原始结构的全局均方根(RMS)偏差程度不同。 (4)使用接近本地的诱饵来测试不同的分子力学和基于知识的能量函数。区分更接近自然结构和更接近自然结构的结构的成功和失败将被用来系统地改进可用于改进接近自然结构的能量函数。

DESCRIPTION (provided by applicant): The overall aim of this proposal is to continue to advance the accuracy of homology modeling for structural genomics. This will be done by advances in two main directions: (a) better classification of protein domains as they are discovered by experimentalists and (b) better methods to refine protein models to bring them closer to the actual structure. Classification of known structures as they are discovered will enable experimentalists to better evaluate their progress and relate it to the work of other groups. It will also provide a valuable data-base of accurate multiple structure alignments for use in both homology modeling and fold recognition. Better refinement will involve the development of energy functions that favor native protein structures at a detailed all-atom level as well as more powerful methods for moving from an initial model towards the real structure. Besides leading to more accurate homology models, both these advances will have far-reaching applications in all studies of molecular function including modeling of ligand binding, modeling of protein-protein interactions and more general simulation of protein function. We believe that such significant progress will be best achieved by successful completion of the following specific aims: (1) Develop a reliable method for domain classification based on our improved structure alignment method Structal, which performed surprisingly well in this study of six different methods. Data will be updated every week using the computer resources that we have available for this grant. (2) Use the regular Structal searches done as part of Aim (1) to produce multiple structural alignments that will be based on our methods for multiple structure superposition and alignment conflict resolution. (3) Use orthogonal normal modes in torsion angle and Cartesian space to generate thousands of near-native decoys. These decoy sets will be generated to have small local deformations but differ from the native structure by different extents of the global root mean square (RMS) deviation. (4) Use near-native decoys to test different molecular mechanics and knowledge-based energy functions. Successes and failures to discriminate structures closer to the native structure from those that are further will be used to systematically improve energy functions that can be used to refine near native structures.