Accurate Molecular Modeling in Structural Genomics

这个提议的总体目标很容易陈述：提供与最好的CASP（结构预测的关键评估）预测相同精度的自动化比较或同源建模。在1998年和2000年的CASP会议上，超过100个小组预测了大约40个目标序列，每个结构的总工作量超过一个人年。有了一个能在几个小时的计算机时间内完成同样工作的编程系统，我们将能够极大地提高结构基因组计划中确定的蛋白质结构的价值。我们的具体工作是：(1)可靠的折叠识别是将预测的目标序列与已知序列和相关已知结构（模板）相匹配的重要的第一步。我们的方法以一种对噪声或任何一种方法的限制不敏感的方式结合了来自最佳可用万维网服务器的结果。通过与圣地亚哥结构基因组学联合中心的Wooley博士和利弗莫尔蛋白质结构预测中心的Fidelis博士的合作，我们可以对我们的方法进行持续的盲测。(2)我们将对我们的结构增强序列比对方法进行校正，以适应改进程序Structal生成的大量精确的结构比对。我们将使用一种新的多结构叠加方法来制作多个序列剖面，以获得更好的比对结果。(3)原子细节的添加是所有同源建模的关键阶段。在这里，我们将使用一种新方法来增强我们经过良好测试的方法SegMod，该方法通过平均场平均来组合来自不同模板结构的数据。(4)不受核磁共振或x射线数据约束的能量最小化通常会破坏结构的构象，而不是使其更像天然结构。这是细化问题，我们的目标是用一种新的方法来解决从高度精细的x射线结构中导出连续的，可微的能量函数。笛卡尔和扭转角最小化将结合起来，以给出最大可能的收敛半径。

The overall aim of this proposal is easily stated: provide automated comparative or homology modeling with the same accuracy as the best CASP (Critical Assessment of Structure Prediction) predictions. At CASP meeting in 1998 and 2000, some 40 target sequences were predicted by over 100 groups, for a total effort of over a man-year per structure. With a programming system that does as well in a few hours of computer time, we will be able to greatly increase the value of protein structures determined in the Structural Genomics Initiative. Our specific efforts are: (1) Reliable fold recognition is the essential first step that matches up the target sequence being predicted with a known sequence and associated known structure (the template). Our method combines the results from the best available world wide web servers in a way that is insensitive to noise or limitations of any one method. Collaborations with Dr. Wooley at the Joint Center for Structural Genomics, San Diego and with Dr. Fidelis at the Protein Structure Prediction Center, Livermore, will allow us to subject our methods to continuous blind testing. (2) We will calibrate our structure enhanced sequence alignment method to fit a large number of accurate structural alignments generated with the improved program, Structal. We will use a new method of multiple structure superposition to make multiple sequence profiles that may give better alignments. (3) Adding atomic detail is a key stage in all homology modeling. Here, we will augment our well-tested method, SegMod, with a new method for combining data from different template structures by mean-field averaging. (4) Energy minimization unconstrained by NMR or x-ray data generally spoils the conformation of a structure rather than making it more native-like. This is the Refinement Problem that we aim to solve using a novel method for deriving continuous, differentiable energy functions from highly refined x-ray structures. Cartesian and torsion angle minimization will be combined to give the largest possible radius of convergence.