METHODS FOR COMPARISON OF PROTEIN THREE DIMENSIONAL STRUCTURE

我们已经开发了快速比较蛋白质结构的算法 以及显著性检验的统计标准。这些将被用来 检测古代进化关系并识别共同的 蛋白质的结构“积木”。我们将两种蛋白质定义为 如果它们具有与次要的相似的成分，则相似 结构元素，如果这些元素在空间中的方向相似， 如果它们以相同的方式在拓扑上相连。我们检测到 使用简化向量的元素的相似方向和拓扑 表示和集团检测算法，由P. 威利特和他的同事。然后识别出相似的子结构 根据它们在比较中出现的概率进行排序 不同的蛋白质；计算假设元素对 比较是独立的，并计算卡方统计 相对于随机元素对分数的经验分布。 这一过程非常迅速，允许将蛋白质与 蛋白质数据库中的非冗余结构约为5 几秒钟。在对照实验中，最显著的“命中”与 总是有一些已知的相似之处。得分最高的元素-路线 然后以这种方式识别作为蒙特卡罗的输入 残基比对过程，该过程使用类似于 这是为蛋白质“穿线”而开发的。比较使用一个 传统的均方根距离度量及其意义 替代可能的路线是相对于经验路线进行评估的 由大小相似的蛋白质随机比对得出的分布 和核心元素组成。确定最令人惊讶的 子结构相似大约需要1分钟，通常会导致 叠加分数高于文献中的分数。这个 这项工作的意义将是将地平线扩展到 可以通过使用结构来检测进化关系 用比较代替单独的序列比较。结果将是 被生物学家作为“Entrez”中的“结构邻居”提供 浏览器。

We have developed algorithms for rapid comparison of protein structure and statistical criteria for significance testing. These will be used to detect ancient evolutionary relationships and to identify the common structural "building blocks" of proteins. We define two proteins to be similar if they have a similar composition with respect to secondary structural elements, if these elements are similarly oriented in space, and if they are topologically connected in the same manner. We detect similar orientation and topology of elements using a simplified vector representation and clique detection algorithm, as suggested by P. Willett and colleagues. The similar substructures identified are then ranked according to the probability that they would occur in comparison of dissimilar proteins; the calculation assumes that element-pair comparisons are independent, and calculates chi-square statistics relative to an empirical distribution of random element-pair scores. This procedure is very rapid, allowing comparison of a protein against the non-redundant structures in the Protein Data Bank in approximately 5 seconds. In control experiments The most significant "hits" correspond invariably to known similarities. The top-scoring element-alignments identified in this way then serve as input to a Monte Carlo residue-alignment procedure, which uses a sampling strategy similar to that developed for protein "threading". Comparison employs a conventional root-mean-square distance metric, and significance of alternative possible alignments is evaluated relative to an empirical distribution derived by random alignment of proteins with similar size and core-element composition. Identification of the most surprising substructure similarity requires roughly 1 minute, and often leads to superposition scores superior to those in the literature. The significance of this work will be in extending the horizon to which evolutionary relationship may be detected, by employing structure comparison in place of sequence comparison alone. The results will be made available to biologists as "structural neighbors" in the "Entrez" browser.