More Power to the Many: Scalable Ensemble-based Simulations and Data Analysis

谷氨酸受体以及了解其结合特性具有重要的生物医学基础，因为它们介导神经元信号传导。该项目旨在表征和理解谷氨酸与 N-甲基-D-天冬氨酸受体 (NMDAr) 的结合，NMDAr 是谷氨酸受体蛋白质家族的成员，对神经科学和药理学具有潜在的深远影响。 然而，NMDAr 配置景观的表征是一个高性能计算 (HPC) 问题。 它需要模拟的时间尺度和系统规模远远超出以前进行的任何模拟。 该项目将利用Blue Waters的千万亿级计算能力，采用新的采样方法以及独创的计算和数据处理技术来研究这样的系统。该项目将使用分子动力学（MD）模拟来研究这种大分子系统。 然而，获得复杂化学系统构型空间的充分采样以准确描述重要亚态的结构特性、它们的相对倾向以及它们之间的可接近的转变仍然是一个挑战。 该项目建议使用一种新颖的软件框架，该框架在正确的计算资源上使我们通过MD对大分子构象空间进行采样的能力发生了重大变化。 该项目将研究一种具有重大生物医学相关性的蛋白质，即 N-甲基-D-天冬氨酸受体 (NMDAr) 的配体结合域 (LBD)，它可以例证这些问题。 软件策略的核心思想与许多其他多尺度方法类似，例如伞式采样、元动力学、自适应偏置方法或过渡路径采样：可以同时模拟许多（数百或数千）短轨迹，而不是运行一个或几个长 MD 轨迹。 使用复杂的数据缩减和分析方法以及粗粒度信息从这些非常大的数据集中提取信息——这体现了理解系统所需的化学洞察力，例如近似自由能——用于细化生成进一步轨迹的方式（即我们采样的方式）。然后，将采样空间的分析结果用于迭代过程，以进一步指导构象空间（即我们采样的位置）的搜索。 Blue Waters 的这项分配将使该项目能够获得总共 2.7 毫秒的 NMDAr LBD 系统模拟。通过我们的方法相对于普通 MD 允许（至少）三个数量级的采样加速，该项目将能够在数秒的时间尺度内绘制与构象动力学相关的该蛋白质的构型景观，即完全表征配体结合结构域在 NMDAr 的生物学功能和机制中的作用。

Glutamate receptors, and understanding their binding characteristics, are of fundamental biomedical importance as they mediate neuronal signaling. This project proposes to characterize and understand glutamate binding to the N-methyl-D-aspartate receptor (NMDAr), a member of the glutamate receptor family of proteins, with potential profound consequences for neuroscience and pharmacology. However, the characterization of the configurational landscape of NMDAr is a High Performance Computing (HPC) problem. It requires simulations with timescales and system sizes well beyond any that have previously been undertaken. The project will use the petascale computing capabilities of Blue Waters to study such a system, using new sampling methods and original computing and data processing techniques.The project will use molecular dynamics (MD) simulations to study this macromolecular system. However, it remains a challenge to obtain an adequate sampling of the configurational space of complex chemical systems to accurately describe the structural properties of important substates, their relative propensities, and accessible transitions between them. The project proposes to use a novel software framework that on the right computational resource makes a step-change in our ability to sample the conformational space of macromolecules by MD. The project will study a protein of great biomedical relevance that exemplifies these issues, namely the ligand binding domain (LBD) of the N-methyl-D-aspartate receptor (NMDAr). The idea at the core of the software strategy is similar to many other multiscale methods -- such as umbrella sampling, metadynamics, adaptive biasing methods, or transition path sampling: instead of one or a few long MD trajectories being run, many (hundreds or thousands) of short trajectories may be simulated concurrently. Information is extracted from these very large datasets using sophisticated data reduction and analysis methods, and the coarse-grained information -- which embodies the chemical insight necessary to understand the system, e.g. an approximate free energy -- is used to refine the way in which further trajectories are generated (i.e., how we sample). Results from the analysis of the space sampled are then used in an iterative process to further direct the search of the conformational space (i.e., where we sample). This Blue Waters allocation will allow the project to access a total of 2.7 milliseconds of simulation of the NMDAr LBD system. With the three orders of magnitude (at least) speed-up in sampling allowed by our methodology with respect to plain MD, the project will be able to map the configurational landscape of this protein relevant for conformational dynamics up to a timescale of seconds, that is, to completely characterize the role of the ligand binding domain in the biological function and mechanism of NMDAr.