Accurate Modeling in Structural Genomics

描述（由申请人提供）：蛋白质分子的三维结构建模具有明显的生物医学重要性，由两股强大的力量驱动。首先是认识到蛋白质凭借其折叠形状在生命系统中执行几乎所有基本的功能和结构任务；几乎所有的药物都依赖于一个小分子通过三维形状互补来抑制功能失调的蛋白质。第二是基因组蛋白序列数据的快速测定，在过去28个月翻了一番，并辅以同样快速的新蛋白结构数据测定；序列的结构覆盖率（包含一些结构信息的序列的百分比）超过50%，并且由于结构基因组学的倡议正在增加。本提案通过开发和改进精确同源建模（已知相关序列的结构）的方法来延续先前的目标。目前的目标扩展到从头算结构预测的一般问题（没有任何相关序列的结构）。这种扩展是可能的，因为最近的进展和认识到同源建模和从头开始结构预测都有一个共同的哲学，植根于我们在1995年开创的诱饵/区别范式。具体来说，蛋白质建模和结构预测都有四个相互关联的阶段：(a)能量函数的制定，(b)移动集的应用，(c)诱饵结构的生成和(d)预测结构的评估。迭代这四个步骤以改进诱饵和能量函数，从而获得更好的预测结构。对实验确定的序列和结构的分析与此计划建模携手并进，以提供问题的范围和该领域取得的进展的概述。在之前的资助期内，我们被这样的分析所吸引，我们希望继续这项活动，特别关注“暗物质”，那些我们信息最少的序列。我们清楚地认识到，这些是雄心勃勃的目标，但最近的进展使我们感到鼓舞。我们的方法使用基于知识或统计的能量函数，但我们的理念是植根于系统的物理本质。因此，我们的工作将对分子功能的理论研究有深远的应用，包括配体结合建模，蛋白质-蛋白质相互作用建模和更一般的蛋白质功能模拟。我们的五个具体目标是：(1)更好的基于知识的能量函数；(2)一般和新颖的移动集；(3)通过均匀采样和强大的搜索生成诱饵；(4)评估结构以揭示缺陷；(5)根据序列域聚类分析未表征序列。实现这些目标将推进我们对分子结构的基本理解：预测的分子结构可以指导实验并导致对分子机制的进一步理解。

DESCRIPTION (provided by applicant): Modeling three-dimensional structure of protein molecules is of clear biomedical importance, driven by two powerful forces. First is the realization that proteins carry out almost all essential functional and structural tasks in living systems by virtue of their folded shape; almost all drugs depend on a small molecule inhibiting a malfunctioning protein through shape complementarity in three-dimensions. Second is the rapid determination of genomic protein sequence data, doubling in the past 28 months, and complemented by equally rapid determination of novel protein structural data; structural coverage of sequences (percentage of sequences with some structural information) is over 50% and is increasing thanks to structural genomics initiatives. This proposal continues previous aims by developing and improving methods for accurate homology modeling (have known structure of a related sequence). Current aims extend to the general problem of ab initio structure prediction (no structure of any related sequence). Such extension is possible due to recent progress and a realization that both homology modeling and ab initio structure prediction share a common philosophy rooted in the decoy / discriminate paradigm we pioneered in 1995. Specifically, both protein modeling and structure prediction have four inter-related stages: (a) Formulation of energy functions, (b) Application of move sets, (c) Generation of decoy structures and (d) Assessment of predicted structures. These four steps are iterated to improve both decoys and energy functions so as to obtain ever better predicted structures. Analysis of experimentally determined sequences and structures goes hand in hand with this planned modeling to give as an over-view of the extent of the problem and the progress made in the field. Drawn to such an analysis in the previous funding period, we expect to continue this activity with particular focus on the 'dark matter', those sequences for which we have least information. We are well-aware that these are ambitious aims but are encouraged by recent progress. Our methodology uses knowledge-based or statistical energy functions, but our philosophy is very rooted in the physical nature of the systems. As such, our work will have far-reaching applications to theoretical studies of molecular function including ligand binding modeling, protein-protein interaction modeling and more general simulation of protein function. Our five specific aims are: (1) Better knowledge-based energy functions, (2) General and novel move sets, (3) Decoy generation by uniform sampling and powerful search and (4) Assessment of structures to reveal deficiencies and (5) Analysis of uncharacterized sequence in terms of clustering sequence domains into new families. Achieving these aims will advance our fundamental understanding of the molecular structure: predicted molecular structure can guide experiments and lead to further understanding of molecular mechanisms. PUBLIC HEALTH RELEVANCE: Modeling three-dimensional structures of protein molecules is of clear biomedical importance: (1) proteins carry out almost all essential functional and structural tasks in living systems by virtue of their folded shape (almost all drugs depend on a small molecule binding to and inhibiting a malfunctioning protein through shape complementarity in three-dimensions); and (2) the rapid growth of genomic protein sequence data, doubling in the past 28 months. This proposal continues previous aims by developing improved methods for accurate homology modeling (have known structure of a related sequence) and also extends the aims to the general problem of ab initio structure prediction (no structure of any related sequence).