CDS&E: Adaptive Biomolecular Free Energy Simulations on Massive Computational Grids

坦普尔大学的Ron Levy获得了化学系化学理论、模型和计算方法计划颁发的奖项，该奖项旨在开发强大的计算和统计工具，以提供对蛋白质如何在控制细胞过程中发挥作用的基本了解。该项目由数学科学部数学与统计科学与工程计算与数据支持计划共同资助，并为国家战略计算倡议做出贡献。要了解蛋白质的功能，需要对控制哪些结构状态具有功能的分子图景进行广泛的调查，并了解与状态之间的转换相关的自由能成本，包括“活跃”和“不活跃”的蛋白质状态。本项目正在开发的基于随机加权的自适应计算技术使得在由数十万个处理器组成的大型计算网格上进行更可靠和更准确的蛋白质-配体结合和蛋白质构象自由能变化的模拟成为可能，并且可以在极大范围内进行。这种能力在制药业和生物技术中有重要的应用。这项研究是在学术环境中进行的，学生们在那里接受最先进的模拟技术培训。这项研究的一个关键方面是创建和传播新的理论和计算方法，以计算机脚本和程序的形式向社会免费提供。本研究的重点是开发新的采样协议，用于在大规模但最少通信的计算网格上执行生物分子模拟。这些算法旨在从拥有约5K处理器的园区网格扩展到拥有约500K处理器的IBM World Community网格。基于未入库的加权直方图分析方程的随机解的新的自由能重加权技术正在形成，当模拟在某些状态处局部平衡但不接近全局平衡时，该技术可用于估计来自多态生物分子模拟的自由能。利维教授正在开发一种自适应版本的异步副本交换，它将自由能重加权方程的随机解直接合并到用于传播轨迹的算法中。该项目解决了使用专门为计算网格设计的多规范抽样方案来执行和分析生物分子模拟所涉及的重大计算和统计挑战，这些抽样方案的大小从数千个处理器(园区网格)到数十万个处理器(IBM世界社区网格(WCG))。两个应用被用来说明新的方法及其在WCG上的实现：设计HIV-1整合酶的抑制剂，以及探索序列对激酶功能转换的影响。利维教授通过美国和欧洲的暑期学校和研讨会，提供有关新代码和技术的讲座和实践经验。代码和脚本可自由传播，供模拟社区使用，并可与GROMACS和OpenMM等广泛使用的开源分子动力学代码接口。

Ron Levy of Temple University is supported by an award from the Chemical Theory, Models and Computational Methods Program in the Chemistry Division for the development of powerful computational and statistical tools to provide a fundamental understanding of how proteins function in controlling cellular processes. This project is co-funded by the Computational and Data-Enabled Science and Engineering in Mathematical and Statistical Sciences Program in the Division of Mathematical Sciences, and contributes to the National Strategic Computing Initiative. Understanding how proteins function requires extensive surveys of the molecular landscapes that control which structural states are functional, and knowing the free energy costs associated with transitions between states, including "active" and "inactive" protein states. The adaptive computational techniques based on stochastic reweighting that are being developed in this project make it possible to perform more reliable and accurate simulations of protein-ligand binding and conformational free energy changes of proteins, and to do so at extreme scale, on large computing grids consisting of hundreds of thousands of processors. This capability has important applications in the pharmaceutical industry and for biotechnology. The research is being carried out in an academic environment, where students are trained in state-of-the-art simulation techniques. A key aspect of the research is the creation and dissemination of the new theoretical and computational approaches, implemented as computer scripts and programs made freely available to the community.The focus of this research is on developing new sampling protocols for performing biomolecular simulations on massive but minimally communicating computational grids. The algorithms are designed to scale from campus grids with ~5K processors to the IBM World Community grid with ~500K processors. New free energy reweighting techniques based on stochastic solutions to the unbinned weighted histogram analysis equations are being formulated which can be used to estimate free energies from multi-state biomolecular simulations, when the simulations are locally equilibrated but not close to globally equilibrated at some of the states. Professor Levy is developing an adaptive version of asynchronous replica exchange that incorporates stochastic solutions to the free energy reweighting equations directly into the algorithms used to propagate the trajectories. This project addresses the significant computational and statistical challenges associated with performing and analyzing biomolecular simulations using multi-canonical sampling schemes designed specifically for use on computational grids, and which span sizes from thousands of processors (campus grids) to hundreds of thousands of processors (the IBM World Community Grid (WCG)). Two applications are used to illustrate the new approach and its implementation on the WCG: designing inhibitors of HIV-1 integrase, and exploring the effect of sequence on functional transitions in kinases. Professor Levy provides lectures and hands-on experience with the new codes and techniques through summer schools and workshops in the U.S. and Europe. Codes and scripts are freely disseminated for use by the simulation community, and can be interfaced to widely-used open source molecular dynamics codes such as GROMACS and OpenMM.