Data-driven Langevin modeling of nonequilibrium processes

如今，全原子分子动力学模拟产生了海量数据，其系统分析构成了一个重大挑战。为了便于对基本物理的简单解释，我们最近发展了一种数据驱动的朗之万方程方法，它从给定的分子动力学轨迹构造低维动力学模型。在研究单元“降低非平衡系统的复杂性”中，该项目将数据驱动的朗之万方程方法扩展到非平衡过程，在非平衡过程中，我们要么从非定常态开始并研究系统到其平衡态的驰豫，要么考虑外部驱动的系统。为此，我们构建了目前研究的各种问题的非平衡朗之万模型，包括各种蛋白质体系中的配体解离和配体诱导的变构转变。针对马尔可夫描述，我们探索了各种措施来优化朗之万场的马尔可夫性。

Nowadays all-atom molecular dynamics simulations produce huge amounts of data, whose systematic analysis poses a major challenge. To facilitate a simple interpretation of the underlying physics, we have recently developed a data-driven Langevin equation approach thatconstructs a low-dimensional dynamical model from a given molecular dynamics trajectory.Within the Research Unit ”Reducing complexity of nonequilibrium systems,” this project extends the data-driven Langevin equationapproach to nonequilibrium processes, where we either start in anonstationary state and study the relaxation of the system into itsequilibrium state, or consider an externally driven system. To thisend, we construct nonequilibrium Langevin models for a variety ofcurrently studied problems, including ligand unbinding andligand-induced allosteric transitions in various proteinsystems. Aiming for a Markovian description, we explore variousmeasures to optimize the Markovianity of the Langevin fields.