云计算-RNA分子力场中静电能项的改进计算.pdf

II

用机器学习的方法预测 RNA 分子的原子多级距可以极大的减少了计算的成本。

本文中采用全连接神经网络的方法对原子多级距进行预测，得到了较好的实验效

果。在改进的实验中，使用剔除 H 原子的结构预测多级距，实验效果得到了较大

的提升，证明了 H 原子的位置信息的冗余性，这说明了 H 原子对 RNA 电子云分

布影响较小可忽略，同时证明了研究过程中使用 H 原子饱和结构的合理性。

III

Force fields are becoming more and more important in scientific research. The

research methods using molecular force field simulation have the advantage of lower

cost and accurate observation of detailed information at each moment in chemical

reaction compared with traditional experiments. RNA is not only a carrier of biological

genetic information, it plays many important roles in the physiology of animals and

humans, and researchers are finding more and more other functions in scientific

experiments.

At present, force fields for RNA typically employ a point charge model of

electrostatics, which does not provide a realistic quantum-mechanical picture, and the

and 15 types of atom-atom interactions. Fragments (phosphate, sugar, base, phosphate-

sugar, sugar-base, nucleotide) are separated and then “capped” according to the

principle of restoring its chemical environment in RNA as much as possible and the

atom-atom electrostatic interaction energy is calculated using atomic multipole

moments. The minimum convergence internuclear distance of the electrostatic energy

for 15 atomic interaction types was obtained. Transferability of the multipole moments

model is also investigated. The experiment proved that the model has good

transferability.

The study of the torsion angle of RNA small molecule fragments is also one of the

important research directions for optimizing force field for RNA molecular. 16 small

IV

molecules cut and capped (according to the principle of restoring its chemical

environment in RNA as much as possible) from 2MVY, and energy minima of each

molecules are obtained. Geometries were optimized at the HF/6-31G(d,p), B3LYP/apc-

1, and MP2/cc-pVDZ levels of theory using Gaussian09. The number of energy minima

is related to the size and the flexibility of the molecule. The torsion angles of α, β, γ, δ,

ε, ξ and χ of RNA were analyzed. Multipole moments of atoms occurring in the common

The works above have proved availability and importance of atomic multipole

moments in force field for RNA molecular. However, using of integral methods to

calculate the atomic multipole moments is expensive, especially when using higher

level of theory. Using machine learning to predict the atomic multipole moments in

RNA molecular can greatly reduce the computational expense. In this work, the method

of fully connected neural network was used for research and prediction of atomic

multipole moments, and good experimental results were obtained. In the improved

experiment, the structure reject hydrogen atom was used to predict the atomic multipole

moments, and the experimental results were greatly improved, which proves the

redundancy of the position information of the H atom, and indicates that the influence

V

1.1 研究背景 .......................................................................................................... 1

1.2 研究现状 .......................................................................................................... 2

1.3 本文的主要工作 .............................................................................................. 5

1.4 本文的结构安排 .............................................................................................. 6

2.1 分子力场 .......................................................................................................... 7

3.1 实验数据 ........................................................................................................ 12

3.1.1 切割与饱和 ............................................................................................. 13

3.2 计算过程和细节 ............................................................................................ 15

3.3 结果和讨论 .................................................................................................... 15

3.3.1 6D 与 L = 10 原子多级距的对比 ............................................................ 17

3.3.2 原子-原子相互作用的最小收敛距离 .................................................... 18

3.3.3 原子多级距等级与核间距关系 ............................................................. 20

3.3.4 不同误差下多级距等级和核间距的关系 ............................................. 22

3.3.5 从 2MVY 到其他 RNA 分子的可转移性 ............................................. 24

4.1 计算过程和计算细节 .................................................................................... 35

4.1.1 能量最小稳定构象 ................................................................................. 35

4.2 结果和讨论 .................................................................................................... 37

4.2.1 扭转角 ..................................................................................................... 37

4.2.2 不同分子的能量最小稳定构象数量 ..................................................... 39

4.2.3 全局能量最小稳定构象的扭转角 ......................................................... 43

4.2.4 不同理论水平的原子多级距 ................................................................. 46

4.3 本章结论 ........................................................................................................ 52

5.1 实验细节 ........................................................................................................ 53

5.1.1 数据集 ..................................................................................................... 53