血红素模型体系多自旋态可变电荷力场开发

Spin crossover is a major feature of two states reaction mechanism, which exists broadly in reactions of transition metal enzymes. At the same time, spin crossover makes a new challenge to force field development of these systems. Two questions should be addressed in the development of the force field for these systems. Firstly, the force field should be able to describe multi spin states simultaneously. Secondly, new methods should be adopted to calculate the switching probability in spin crossover region. Currently, force field that fit both of these features is still rarely reported. In this project, we proposed to development of multi spin states updateable charge force filed for heme models and apply the force field to multi spin states molecular dynamic simulation. At the first place, we use data driving model to build multi spin states updateable charge force filed. The trajectory of on-the-fly molecular dynamic simulation would be clustered into several classes and each class is represented by a structure. The multi spin states parameters of interested structures would, then, be calculated by a weighted average method. Meanwhile, the modified Zhu-Nakamura method is introduced into molecular dynamic simulation for calculating the switching probability in spin crossover region. This algorithm requires only energy and grad as input, both of which are outputs from force field. The multi spin states molecular dynamic simulations will be performed in the project and with heme model system as test. This strategy is also helpful to the developments of force fields for other transition metal enzymes.

自旋交叉在含过渡金属的酶催化反应中广泛存在，是两态反应体系的重要特征，给这些体系的力场开发提出了新的挑战。主要表现在两个方面，一是需要力场能够同时描述多个自旋态，二是在自旋交叉区域，需要根据力场直接计算跃迁几率。目前，同时满足这两个要求的力场鲜有报导。本项目拟开发多自旋态可变电荷力场，以血红素模型体系为代表实现多自旋态的分子动力学模拟。首先，为了使力场能够描述多自旋态，拟从数据驱动的思想出发，采用从头算动力学方法采样构建结构数据集，通过聚类、网络等方法对数据集和轨迹进行分析获得代表结构集，采用代表结构加权的方法实时预测各自旋态电荷，开发多自旋态可变电荷力场。其次，在分子动力学中引入Zhu-Nakamura非绝热跃迁算法处理自旋交叉过程，这一算法只需要力场提供能量和梯度作为输入，可实现血红素模型体系多自旋态的分子动力学模拟。本项目研究思路具有很好的可拓展性，可为其它类似力场的研究提供借鉴。

自旋交叉在含过渡金属的酶催化反应中广泛存在，是两态反应体系的重要特征，给这些体 系的力场开发提出了新的挑战。主要表现为需要力场能够同时描述多个自旋态和在自旋交叉区域，需要根据力场直接计算跃迁几率。本项目开发了多自旋态可变电荷力场，以血红素模型体系为代表开发了多自旋态的分子动力学模拟程序，在分子动力学中引入Zhu-Nakamura非绝热跃迁算法处理自旋交叉过程，这一算法只需要力场提供能量和梯度作为输入，可实现血红素模型体系多自旋态的分子动力学模拟。. 项目研究中，以血红素体系为代表，基于从头算动力学方法产生代表性结构，优化了数据集构建方法，发展了ATSF描述符，采用随机森林方法开发了多自旋组态可变电荷力场; 基于多自旋态可变电荷力场的思想，将机器学习力场方法进一步用于糖类分子的异构势能面计算，在力场精度、偶极矩方面比传统力场有提升，特别是偶极矩在没有经过学习就可以预测，说明力场具有一定的可解释性。通过水合自由能和糖类分子与蛋白相互作用能测试，力场表现均较传统力场有提升; 以偶氮苯分子为例，测试了可变电荷力场在光致顺反异构分子基态和激发态力场的表现，对异构化势能面的分析表明，直接预测电荷精度不足，需要预测激发态的能量和梯度。采用Python程序的PY4MPI结构，组合Tinker程序和机器学习力场预测程序，实现了多自旋态自旋交叉区域的模拟。