USING THE FOLDING PROCESS TO IMPROVE PROTEIN STRUCTURE PREDICTION

这个子项目是许多研究子项目中利用 资源由NIH/NCRR资助的中心拨款提供。子项目和 调查员(PI)可能从NIH的另一个来源获得了主要资金， 并因此可以在其他清晰的条目中表示。列出的机构是 该中心不一定是调查人员的机构。 确定蛋白质结构的计算方法对基因组计划至关重要，因为大量新的蛋白质序列的性质完全未知。目前蛋白质结构预测算法的一个主要局限性是由两个主要的二级结构服务器产生的输入二级结构预测的质量不够高。这些服务器使用广泛依赖于序列相似性(称为同源性)的机器学习方法，因此，尽管知道三级上下文经常影响二级结构，但仍使用序列局部信息。我们设计了一种蒙特卡罗模拟退火法和一套相应的计算机程序来实现一种新的方案，在该方案中，二级结构和三级结构都是以自洽的Bootstrap方式预测的，而不使用同源信息。使用TeraGrid开发拨款进行的测试表明，我们的方法在二级结构预测方面优于领先的服务器，并提供与最佳方法相当的三级结构(使用的计算机时间少两个数量级！)对于小的(少于120个残基)单域蛋白。这一建议旨在改进和扩展我们的预测方法，并提高它们的计算效率。建议的项目包括使用序列相似性来改进我们的移动集，引入二级结构分配的动态标准，该标准取决于先前在模拟过程中分配的结构的比例，改进能量函数以提高预测质量并能够处理更大的蛋白质等。广泛的应用将考虑具有不寻常或困难的三级结构的广泛的蛋白质。

This subproject is one of many research subprojects utilizing the resources provided by a Center grant funded by NIH/NCRR. The subproject and investigator (PI) may have received primary funding from another NIH source, and thus could be represented in other CRISP entries. The institution listed is for the Center, which is not necessarily the institution for the investigator. Computational methods for determining protein structure are essential to the genome project due to the huge number of new sequences for proteins whose properties are completely unknown. A major limitation to current protein structure prediction algorithms is the inadequate quality of the input secondary structure predictions produced by the two major secondary strcture servers. These servers use machine learning methods that extensively rely on sequence similarity (called homology) and, hence, use sequence local information despite the knowledge that tertiary context often influences the secondary structure. We have devised a Monte Carlo simulated annealing algorithm and a corresponding set of computer codes to implement a novel scheme in which both secondary and tertiary structure are predicted in a self-consistent bootstrap fashion without the use of homology information. Tests made using a teragrid development grant demonstrate that our method outperforms the leading servers in secondary structure prediction and provides comparable tertiary structures to the best methods (using two orders of magnitude less computer time!) for small (less than 120 residues) single domain proteins. This proposal seeks to improve and extend our predictive methods as well as increase their computational efficiency. Proposed projects include the use of sequence similarity to improve our move set, the introduction of dynamic criteria for secondary structure assignment that vary depending of the fraction of structure previously assigned during the simulations, the improvement of the energy function to enhance the predictive quality and enable treating larger proteins, etc. Extensive applications will consider a wide range of proteins with unusual or difficult tertiary structures.