College of Biological Sciences, University of Tsukuba, 1-1-1, Tennodai, Tsukuba, Ibaraki 305-0821, Japan.
Center for Computational Sciences, University of Tsukuba, 1-1-1 Tennodai, Tsukuba, Ibaraki 305-8577, Japan.
J Chem Inf Model. 2020 Aug 24;60(8):4021-4029. doi: 10.1021/acs.jcim.0c00580. Epub 2020 Aug 13.
Molecular dynamics (MD) simulation has become a powerful tool because it provides a time series of protein dynamics at high temporal-spatial resolution. However, the accessible timescales of MD simulation are shorter than those of the biologically rare events. Generally, long-time MD simulations over microseconds are required to detect the rare events. Therefore, it is desirable to develop rare-event-sampling methods. For a rare-event-sampling method, we have developed parallel cascade selection MD (PaCS-MD). PaCS-MD generates transition pathways from a given source structure to a target structure by repeating short-time MD simulations. The key point in PaCS-MD is how to select reasonable candidates (protein configurations) with high potentials to make transitions toward the target structure. In the present study, based on principal component analysis (PCA), we propose PCA-based PaCS-MD to detect rare events of collective motions of a given protein. Here, the PCA-based PaCS-MD is composed of the following two steps. At first, as a preliminary run, PCA is performed using an MD trajectory from the target structure to define a principal coordinate (PC) subspace for describing the collective motions of interest. PCA provides principal modes as eigenvectors to project a protein configuration onto the PC subspace. Then, as a production run, all the snapshots of short-time MD simulations are ranked by inner products (IPs), where an IP is defined between a snapshot and the target structure. Then, snapshots with higher values of the IP are selected as reasonable candidates, and short-time MD simulations are independently restarted from them. By referring to the values of the IP, the PCA-based PaCS-MD repeats the short-time MD simulations from the reasonable candidates that are highly correlated with the target structure. As a demonstration, we applied the PCA-based PaCS-MD to adenylate kinase and detected its large-amplitude (open-closed) transition with a nanosecond-order computational cost.
分子动力学（MD）模拟已成为一种强大的工具，因为它可以提供高时空分辨率的蛋白质动力学时间序列。然而，MD 模拟的可访问时间尺度比生物罕见事件的时间尺度短。通常，需要进行微秒级别的长时间 MD 模拟才能检测到罕见事件。因此，开发罕见事件采样方法是可取的。对于罕见事件采样方法，我们已经开发了并行级联选择 MD（PaCS-MD）。PaCS-MD 通过重复短时间 MD 模拟，从给定的源结构生成到目标结构的跃迁途径。在 PaCS-MD 中，关键是如何选择具有高势能的合理候选者（蛋白质构型），以朝着目标结构进行跃迁。在本研究中，基于主成分分析（PCA），我们提出了基于 PCA 的 PaCS-MD 来检测给定蛋白质的集体运动的罕见事件。这里，基于 PCA 的 PaCS-MD 由以下两个步骤组成。首先，作为初步运行，使用来自目标结构的 MD 轨迹执行 PCA，以定义描述感兴趣的集体运动的主坐标（PC）子空间。PCA 提供主模式作为特征向量，将蛋白质构型投影到 PC 子空间上。然后，作为生产运行，所有短时间 MD 模拟的快照通过内积（IP）进行排序，其中 IP 定义为快照和目标结构之间的内积。然后，选择具有更高 IP 值的快照作为合理的候选者，并从它们独立重启短时间 MD 模拟。通过参考 IP 的值，基于 PCA 的 PaCS-MD 从与目标结构高度相关的合理候选者中重复短时间 MD 模拟。作为演示，我们将基于 PCA 的 PaCS-MD 应用于腺苷酸激酶，并以纳秒级别的计算成本检测到其大振幅（开-闭）跃迁。
