STRUCTURE PREDICTION BY PROTEIN THREADING

我们开发了计算机方法来预测蛋白质的三维 通过识别结构数据库中类似的“折叠”来构建结构。 蛋白质的序列通过替代构象“串联”，并且 那些最相容的由近似自由能确定 计算，使用接触电位。 研究主要集中在三个方面 领域：1) 使用新颖序列测试线程方法，2) 开发用于预测的置信度统计数据，以及 3) 开发 改善接触电位。 “自适应”线程算法， 基于块对齐和吉布采样优化，在 阿西洛玛研讨会关于该主题的盲预测实验。 在两个预测中的两个准确地表示了未知数 在我们的“核心”数据库中，正确的模型被确定为 top 3，具有精确的序列结构比对。 在另外两个 预测测试统计数据表明信心不足，并且这些蛋白质 被证明与已知的折叠具有不太广泛的相似性，但不准确 表示在核心数据库中。 为这些开发的测试统计数据 预测实验基于打乱的分数分布 序列，这个程序使我们能够严格纠正由于偏差 序列组成和对齐-空间大小。 预测 实验还表明，潜力和核心能力的适度提高 定义将被证明很重要，因为一些预测正好低于 置信度阈值。 为此我们已经开始开发 新的势，基于接触对的更详细的表示 几何形状，以及保守核心子结构的更精确定义。 这项研究的意义在于它可能允许3维 仅与已知蛋白质有较远关系的序列建模 结构，从而提出关于其作用机制的假设 和功能。 我们对肥胖基因产物的线程预测，对于 例如，有人认为与遗传性肥胖有关的瘦素是 结构上和功能上可能与螺旋细胞因子相似。

We have developed computer methods to predict protein three-dimensional structure by recognition of similar "folds" in the structural database. A protein's sequence is "threaded" through alternative conformations, and those most compatible are identified by approximate free-energy calculation, using contact potentials. Research has focussed in three areas: 1) testing of the threading method with novel sequences, 2) development of confidence statistics for predictions, and 3) development of improved contact potentials. The "adaptive" threading algorithm, based on block alignment and Gibb's sampling optimization, was tested in blind prediction experiments for the Asilomar workshop on this topic. In two of two predictions where the unknown was accurately represented in our "core" database, the correct model was identified as one of the top 3, with accurate sequence-structure alignment. In two other predictions test statistics indicated poor confidence, and these proteins proved to have less extensive similarity to known folds, not accurately represented in the core database. Test statistics developed for these prediction experiments were based on score distributions for shuffled sequences, a procedure which allow us to rigorously correct for bias due to sequence composition and alignment -space size. Prediction experiments also suggested that modest improvement in potentials and core definition will prove important, since some predictions were just below the confidence threshold. For this reason we have begun development of new potentials, based on more detailed representation of contact-pair geometry, and more precise definitions of conserved core substructure. The significance of this research is that it may allow 3-dimensional modeling for sequences only distantly related to proteins of known structure, and thus suggest hypotheses as to their mechanism of action and function. Our threading prediction for the Obese gene product, for example, has suggested that leptin, implicated in hereditary obesity, is structurally and perhaps functionally similar to the helical cytokines.