Highly multidimensional thermodynamic property prediction for chemical design using atomistic simulations

弗吉尼亚大学的Michael Shirts得到了化学理论、模型和计算方法计划颁发的奖项的支持，该计划旨在开发方法来有效地预测大量化学物种的热力学性质，并使用先进的统计和模拟技术来描述这些物种的替代模型。通过物理模型预测给定化学体系的热力学性质本质上是一个统计估计问题，但许多潜在有用的统计技术目前还没有用于分子模拟。该项目首次将现有的几种统计技术应用于热力学性质估计问题，并探索了几种新的模拟技术。最重要的是，该项目将现有技术从少数几个状态采样和计算热力学性质的能力扩展到大规模化学性质预测和设计所需的指数级大的多维空间。改进的优化化学空间热力学性质的能力将对药物和材料设计产生影响。这种能力将使探索新的蛋白质、结构杂多聚合物和其他化学复杂的多相体系的性质变得容易得多，潜在地揭示了以其他方式无法发现的新的结构和功能框架。例如，拟议的技术将使优化小分子成为可能，使其在给定病毒蛋白的所有单点突变上具有最紧密的结合亲和力。它还将使研究人员能够快速确定哪些拟议的分子模型最能描述自然化学体系。该方案中的方法将在GROMACS中实现，这是一个广泛使用的开源分子动力学软件工具包，使大量研究人员能够使用它们。这些发现和工具将被合并到AlChemistry.org上，这是一个分享经典分子体系中自由能计算技术、例子和方法的门户网站。

Michael Shirts of the University of Virginia is supported by an award from the Chemical Theory, Models and Computational Methods program to develop methods to efficiently predict the thermodynamic properties of large numbers of chemical species and alternative models to describe these species using advanced statistical and simulation techniques. Predicting thermodynamic properties for a given chemical system via physical modeling is fundamentally a statistical estimation problem, but many potentially useful statistical techniques are not currently used in molecular simulation. This project applies several existing statistical techniques to the problem of thermodynamic property estimation for the first time and explores several novel simulation techniques. Most significantly, the project expands the ability of existing techniques for sampling from and computing thermodynamic properties of a few states to the exponentially large multidimensional spaces required for large-scale chemical property prediction and design.Improved capabilities to optimize thermodynamic properties over chemical space would have impact in drug and materials design. Such capabilities would make exploring the properties of new proteins, structured heteropolymers, and other chemically complex heterogeneous systems much easier, potentially revealing new structural and functional frameworks that could not otherwise be discovered. For example, the proposed techniques would make it possible to optimize small molecules to have the tightest possible binding affinity across all single-point mutations of a given viral protein. It would also allow researchers to rapidly identify which proposed molecular models best describe natural chemical systems. The methods in this proposal will be implemented in GROMACS, a widely-used open source package of molecular dynamics software tools, making them accessible to large numbers of researchers. The findings and tools will be incorporated into Alchemistry.org, a web portal for sharing techniques, examples, and methods for free energy computations in classical molecular systems.