分子动力学模拟和机器学习结合在药物设计领域的应用

Application of Molecular Dynamics Simulations Meet Machine Learning in Structure-based Drug Design

关键词:  药物设计  分子动力学模拟  机器学习

Key words:  drug design  molecule dynamics simulation  machine learning

摘要:

近年来，传统计算机辅助药物分子设计(computer-aided drug molecular design，CADD)和人工智能技术的融合为创新药物发现带来新的契机。分子动力学模拟(molecular dynamics simulation，MD)是CADD中一项不可或缺的经典技术，作为一种强大的模拟计算方法，常常用来研究靶标和药物之间相互作用的动态过程；机器学习(machine learning，ML)作为一种数据驱动的建模方法，正在被广泛应用于药物发现的各个阶段。MD和ML内在互补的特性，为两者的有机结合提供了诸多可能。一方面，ML可以用来分析MD中产生的海量、高维结构动态信息，通过特征提取、降维等策略，识别关键状态和构象，阐明生物体系动态演变背后的潜在机制；另一方面，MD产生的包含结构动态信息的数据可以用于ML模型训练，提高模型对靶标-药物体系热力学性质和动力学性质的预测精度。因此，MD和ML的有效结合在基于结构的药物设计领域的应用具有重要意义。

Abstract:

In recent years, the rise and development of computer-aided drug molecular design(CADD) together with artificial intelligence has greatly accelerated the research paradigm shift in drug discovery. As an indispensable technology in CADD, molecular dynamics simulation(MD) is often used to study the dynamic process of protein-ligand complexes; as a data-driven modeling method, machine learning(ML) has been widely used in various stages of drug discovery. The intrinsic complimentary properties of MD and ML bring up a number of possibilities for their integration. On the one hand, ML can be employed to analyze massive and high-dimensional information generated by MD, revealing key conformations and states through strategies such as feature extraction and dimensionality reduction to elucidate the underlying molecular mechanisms of biological systems. On the other hand, the MD data comprising dynamic information can be utilized to train the ML models and improve their accuracy in predicting thermodynamic and kinetic properties of protein-ligand systems. Therefore, the application of the combination of MD and ML in the field of drug design is of great significance.