STRUCTURE PREDICTION BY PROTEIN THREADING

我们已经开发了计算机方法来预测蛋白质的三维结构， 通过识别折叠基序来构造。蛋白质的序列是 通过数据库中的替代主干结构进行“线程化”，以及 鉴定出由该序列最稳定的构象。 的 工作集中在三个方面：1）快速评估的衍生 经验自由能函数，2）快速算法测试 穿过有缺口的“核心”基序，以及3）核心的识别 子结构将被包括在数据库中。今年我们展示了 基于残基接触电势的能量函数可以识别 正确的核心基序和数十亿之间的对齐 替代品，一个特异性足以为许多预测问题。 我们 已经开发出了序列效应的统计校正 组成和长度，否则会降低基序识别 的特异性 我们已经测试了一个快速线程算法的基础上蒙特 卡罗技术，并发现它是能够确定最佳的 序列和固定大小的核心基序的比对在几秒钟内， 在所有情况下，该最佳值可以通过以下方式确定： 枚举 为了定义一个折叠图案的数据库， 从蛋白质数据库亚结构组成的主要螺旋和 β-链，使用抗局部扭曲的算法和/或 坐标不精确 每一个都定义了一个相关的核心基序家族， 当用允许删除和/或 单个二级结构元件的大小变化。 测试 该自适应线程算法及其相关数据库在 中求进工作总这项研究的意义在于，这些方法可以 允许自动搜索折叠基序数据库，预测 蛋白质中的亚结构构象可能很少或没有 与晶体学数据库中的蛋白质同源。

We have developed computer methods to predict protein three-dimensional structure by recognition of folding motif. A protein's sequence is "threaded" through alternative backbone structures in a database, and the conformations most stabilized by that sequence are identified. The work has focused on three areas: 1) derivation of a rapidly-evaluated empirical free energy function, 2) testing of algorithms for fast threading through gapped "core" motifs, and 3) identification of core sub-structures to be included in the database. This year we have shown that an energy function based on residue contact potentials can identify the correct core motif and alignment among many billions of alternatives, a specificity sufficient for many prediction problems. We have developed statistical corrections for effects of sequence composition and length, which otherwise reduce motif-recognition specificity. We have tested a fast-threading algorithm based on a monte carlo technique, and found that it is capable of identifying optimal alignments of sequence and fixed-size core motifs in a few seconds, in all cases where this optimum can be determined with certainty by enumeration. To define a database of folding motifs we have extracted from the Protein Data Bank substructures consisting of major helices and beta-strands, using an algorithm resistant to local distortions and/or coordinate imprecision. Each defines a family of related core motifs, when threaded with a monte carlo algorithm that allows deletions and/or changes in size of individual secondary structure elements. Testing of this adaptive threading algorithm and its associated database is in progress. The significance of this research is that these methods may allow automated search of a folding motif database, predicting substructure conformations in proteins which may share little or no homology with proteins in the crystallographic database.