Narang Pooja, Bhushan Kumkum, Bose Surojit, Jayaram B
Department of Chemistry, Indian Institute of Technology, Hauz Khas, New Delhi 110016, India.
Phys Chem Chem Phys. 2005 Jun 7;7(11):2364-75. doi: 10.1039/b502226f.
Impressive advances in the applications of bioinformatics for protein structure prediction coupled with growing structural databases on one hand and the insurmountable time-scale problem with ab initio computational methods on the other continue to raise doubts whether a computational solution to the protein folding problem--categorized as an NP-hard problem--is within reach in the near future. Combining some specially designed biophysical filters and vector algebra tools with ab initio methods, we present here a promising computational pathway for bracketing native-like structures of small alpha helical globular proteins departing from secondary structural information. The automated protocol is initiated by generating multiple structures around the loops between secondary structural elements. A set of knowledge-based biophysical filters namely persistence length and radius of gyration, developed and calibrated on approximately 1000 globular proteins, is introduced to screen the trial structures to filter out improbable candidates for the native and reduce the size of the library of probable structures. The ensemble so generated encompasses a few structures with native-like topology. Monte Carlo optimizations of the loop dihedrals are then carried out to remove steric clashes. The resultant structures are energy minimized and ranked according to a scoring function tested previously on a series of decoy sets vis-a-vis their corresponding natives. We find that the 100 lowest energy structures culled from the ensemble of energy optimized trial structures comprise at least a few to within 3-5 angstroms of the native. Thus the formidable "needle in a haystack" problem is narrowed down to finding an optimal solution amongst a computationally tractable number of alternatives. Encouraging results obtained on twelve small alpha helical globular proteins with the above outlined pathway are presented and discussed.
一方面，生物信息学在蛋白质结构预测应用方面取得了令人瞩目的进展，同时结构数据库不断增加；另一方面，从头计算方法存在难以克服的时间尺度问题，这使得人们持续怀疑，对于被归类为NP难问题的蛋白质折叠问题，在不久的将来是否能够通过计算得到解决。我们将一些专门设计的生物物理过滤器和向量代数工具与从头计算方法相结合，在此展示了一条有前景的计算途径，用于从二级结构信息出发确定小型α螺旋球状蛋白质的类天然结构。该自动化方案通过在二级结构元件之间的环周围生成多个结构来启动。引入了一组基于知识的生物物理过滤器，即持久长度和回转半径，它们是在大约1000个球状蛋白质上开发和校准的，用于筛选试验结构，以滤除不可能是天然结构的候选者，并减小可能结构库的规模。如此生成的集合包含一些具有类天然拓扑结构的结构。然后对环二面角进行蒙特卡罗优化以消除空间冲突。对所得结构进行能量最小化，并根据先前在一系列诱饵集与其相应天然结构上测试的评分函数进行排序。我们发现，从能量优化的试验结构集合中挑选出的100个最低能量结构中，至少有几个与天然结构的距离在3 - 5埃以内。因此，令人生畏的“大海捞针”问题被缩小到在计算上易于处理的数量的替代方案中找到最优解。本文展示并讨论了使用上述途径在十二个小型α螺旋球状蛋白质上获得的令人鼓舞的结果。
