Elastic Shape Analysis for Protein Structure Alignment-New Advancement in an Old

描述（由申请人提供）：蛋白质的结构比较对于理解蛋白质之间的进化关系，预测蛋白质结构和预测蛋白质功能非常重要。尽管过去多年来进行了许多研究，但它仍然是一个具有挑战性和开放性的问题。蛋白质是灵活的分子，其结构的严格匹配，正如目前大多数方法所使用的那样，很难识别相对遥远的、功能上重要的相似性。结构比较中另一个众所周知的问题是缺乏全面的统计框架来评估单个蛋白质结构和蛋白质结构类别之间相似性的统计意义。在这个提议中，我们开发了基于弹性形状分析（ESA）的蛋白质结构比较和比对方法。ESA允许灵活匹配，使用两个蛋白质结构的拉伸和弯曲的组合，通过形式距离，测地线距离来量化。通过高效的动态规划算法，可以求出任意两个结构之间最优匹配所对应的最小测地线距离。可以计算一组结构的均值和协方差。可以制定严格的统计框架进行结构比较和分类。在这个框架下，可以评估两个结构之间的相似性；蛋白质家族内的家族特异性结构变异可以表征；并对结构分类进行假设检验。在此基础上，我们提出：1)利用蛋白质结构家族构建的概率分布，建立统一的蛋白质结构分类统计框架；2)基于一组蛋白质结构计算的平均结构，发展了多结构比对方法；3)提出了一种结合序列-结构空间的蛋白质结构比对方法，将主链几何信息和序列信息结合到结构比对中。

DESCRIPTION (provided by applicant): Structure comparison of proteins is important for understanding the evolutionary relationships between proteins, predicting protein structures and predicting protein functions. Despite many studies in the past over the years, it remains a challenging and open problem. Proteins are flexible molecules and rigid matching of their structures, as used by most current methods, has the difficulty of recognizing relatively distant, functionally important similarities. Another well known issue in structure comparison is the lack of comprehensive statistical framework for assessing the statistical significance of similarities between individual protein structures and classes of protein structures. In this proposal, we develop methods based on elastic shape analysis (ESA) for protein structure comparison and alignment. ESA allows flexible matching using a combination of stretching and bending of two protein structures, which is quantified by a formal distance, geodesic distance. The minimum geodesic distance, corresponding to the best matching between any two structures, can be obtained by efficient dynamic programming algorithm. Mean and covariance of a population of structures can be calculated. Rigorous statistical framework can be developed for structure comparison and classification. Under this framework, similarities between two structures can be assessed; family-specific structure variations within a protein family can be characterized; and hypothesis testing for structure classification can be conducted. Based on the framework, we propose to 1) develop a unified statistical framework for classification of protein structures usin probability distributions built from families of protein structures; 2) develop multiple structure alignment method based on the mean structure calculated for a group of protein structures; and 3) develop a method for aligning protein structures on the joint sequence-structure space to incorporate both backbone geometric and sequence information into structure alignment.