New Methods for Large-scale Computer Simulation

项目摘要 仅从蛋白质的一级序列预测其三维结构仍然是一个困难的问题。 这是现代计算生物学中一个宏大而又难以捉摸的挑战。分子动力学模拟具有较高的 有望在分子细节上预测蛋白质结构和折叠途径。最新研究进展 经过验证的计算机硬件和增强的采样方法使从头开始折叠蛋白质成为可能， 大号的。改进后的计算机硬件的亮点是安东，一个大规模并行的专用 D. E.设计的超级计算机。Shaw Research. Anton成功地折叠了D14 A快速折叠突变体， 80个残基的l-阻遏物，在643μ s长的模拟中在49微秒（μs）实现。上 另一方面，增强采样方法的最新进展是由单拷贝连续 模拟回火（CST）方法开发的PI的小组。Klaus Schulten incorpo博士的团队- 将CST方法加入NAMD程序包，重复折叠80个残基的l-阻遏物HG突变体 在10μ s长的模拟中，在0.5和4μs时从完全伸展构象到天然状态， 均方偏差（Ca-RMSD）为1.7 μ m。与此形成鲜明对比的是， 即使在100μs内，使用Anton在多个温度下也没有观察到相同蛋白质的完全折叠。 长时间的模拟CST在折叠模拟中的这种性能从未被任何其他同类产品所匹配。 在传统的计算平台上用于类似目的的方法。最近，为了进一步加强 为了提高研究大型系统的采样效率，PI开发了一种功能更强大的并行CST（PCST） 法trp-笼的初始从头折叠模拟清楚地表明，PCST的效率， 多重折叠和解折叠事件的产生甚至大大上级CST。PCST方法 为三个具体目标的拟议研究奠定了坚实的基础：1）。PCST的发展 增强采样方法; 2）.改进的温度相关约束方案设计 有针对性的抽样; 3）.改进的盲模型选择方法用于有效的目标选择 选举。我们深入的初步研究表明，这些新方法明显优于现有的所有方法- 的方法，并提出了一个高的承诺，成功的拟议中的研究。最终，这些强大的 新的算法将为蛋白质模拟提供急需的工具，并提供有效的解决方案， 在实验X射线晶体学和电子冷冻显微镜中的结构细化。

Project Summary To predict the three-dimensional structures of a protein solely from its primary sequence remains a grand and elusive challenge in modern computational biology. Molecular dynamics simulation has a high promise for predicting protein structures and folding pathways at molecular details. Recent advances in im- proved computer hardware and enhanced sampling methods have made it possible to ab initio fold proteins of larger size. The highlight of the improved computer hardware is Anton, a massively parallel special-purpose supercomputer designed by D.E. Shaw Research. Anton successfully folded the D14A fast-folding mutant of the 80-residue l-repressor, which was achieved at 49 microseconds (μs) in 643μs-long simulations. On the other hand, the latest advance in enhanced sampling methods is represented by the single-copy continuous simulated tempering (CST) method developed by the PI’s group. The group of Dr. Klaus Schulten incorpo- rated the CST method into the NAMD package, which repeatedly folded the 80-residue l-repressor HG mutant from a fully extended conformation to the native state at 0.5 and 4μs in 10μs-long simulations with Ca root- mean-square deviations (Ca-RMSD) of 1.7 Å on a conventional computing platform. In marked contrast, a complete folding of the same protein was NOT observed using Anton at multiple temperatures even in 100μs- long simulations. This performance of CST in folding simulation has never been matched by any other sam- pling method for similar purposes on conventional computing platforms. Most recently, to further enhance sampling efficiencies in studying larger systems, the PI has developed a more powerful parallel CST (PCST) method. Initial ab initio folding simulation of trp-cage clearly demonstrated that the efficiency of PCST in facili- tating multiple folding and unfolding events was even drastically superior to that of CST. The PCST method serves as a solid foundation for the proposed research in three Specific Aims: 1). Development of the PCST method for enhanced sampling; 2). Design of advanced temperature-dependent restraint schemes for targeted sampling; 3). Development of advanced blind model selection methods for efficient target se- lection. Our in-depth preliminary studies demonstrate that these new methods clearly outperformed all exist- ing methods and suggest a high promise of success for the proposed research. Ultimately, these powerful new algorithms will provide urgently-needed tools for protein simulations, and offer an effective solution for structural refinement in experimental X-ray crystallography and electron cryo-microscopy.