Surface Shape Based Screening of Large Protein Databases

描述(申请人提供)：在结构基因组学时代，需要提取和表示蛋白质结合部位的3D形状，以便以健壮和简单的方式重复使用信息。这一建议的长期目标是开发一套计算算法和一个数据库，用于使用蛋白质的局部表面形状特征来预测蛋白质的功能，并用于蛋白质-蛋白质对接预测。将预先计算的结合位点存储在数据库中可实现快速筛选和比较。将开发识别、表示、比较、聚类和对接蛋白质局部表面形状特征的算法。为了识别结合部位，将使用基于可见性的算法，该算法可以识别空洞和突起区域。为了表示识别的结合位点，提出了三个层次的表示。最简单的表示方案使用捕捉全局或局部最大/最小平均曲率、结合部位的半径和深度的特征点。第二级表示使用基于直方图的方法，捕捉结合位点的特征点之间的相对距离。最后一种表示采用了体素化方法。结合位点数据库将采用基于R-树的多维索引，以允许实时筛选和聚集结合位点。它还计划使用自组织映射进行集群，允许动态更新集群。通过R树或自组织映射预先计算的分层聚类提供了由可缩放用户界面显示的动态聚类的框架。提出了一种基于几何哈希的蛋白质-蛋白质快速对接算法，该算法利用预先计算的结合位点来减少搜索空间。在散列步骤中将使用新的不变基，将算法的复杂度从O(N3)降低到O(N2)。在散列中使用了一种新的非均匀哈希表，该表对小错误或参数变化具有容错性。该方法将扩展到能够处理可能存在误差的预测结构。活性位点鉴定方法将被应用于结构基因组计划确定的未知功能蛋白质结构的功能预测。该对接算法将应用于大肠杆菌的蛋白质-蛋白质相互作用数据。

DESCRIPTION (provided by applicant): In the structural genomics era, there is a need to extract and represent 3D shapes of protein binding sites, in order to reuse the information in a robust and simple manner. The long term objective of this proposal is the development of a set of computational algorithms and a database for using local surface shape signatures of proteins to predict function of proteins, and for protein-protein docking prediction. Storing precalculated binding sites in a database allows fast screening and comparison. Algorithms for identifying, representing, comparing, clustering, and docking local surface shape signatures of proteins will be developed. To identify binding sites, a visibility based algorithm will be used, which can identify both cavity and protrusion regions. To represent identified binding sites, three hierarchical levels of representation are proposed. The simplest representation scheme uses feature points which capture global or local maximum/minimum mean curvatures, the radius, and depth of a binding site. The second level of the representation uses a histogram-based method, capturing relative distances between feature points of a binding site. The last representation employs a voxelization method. The database of binding sites will employ R-tree based multidimensional indexes to allow real-time screening and clustering of binding sites. It is also planned to use a Self-Organizing Map for clustering, which allows dynamic updating of clusters. Pre-calculated hierarchical clusters by the R-tree or a Self-Organizing Map provide a framework for dynamic clustering displayed by a zoomable user interface. The fast geometric hashing-based protein-protein docking algorithm is developed, which uses precalculated binding sites to reduce the search space. A new invariant basis will be used in the hashing step to reduce the complexity of the algorithm from O(n3) to O(n2). A novel non-uniform hashing table is used in the hashing, which is tolerant to small errors or changes of parameters. The methodology will be extended to be able to handle predicted structures with possible errors. The active site identification methods will be applied to predict function of protein structures of unknown function determined by structural genomics projects. The docking algorithm will be applied to protein-protein interaction data of E.coli.