Abstract: Predicting the binding affinity between drug molecules and target proteins contributes to understanding biological systems and aiding drug discovery. With the development of computational biology technology driven by biological big data, the research strategy of the binding affinity between drug molecules and target proteins moves from traditional single biomedical experiments into integrated computational technology-aided prediction, which provides new technologies and methods for drug discovery. Since it is important to investigate the binding affinity between drug molecules and target proteins, this paper reviews the advances of it from two aspects, including traditional biological experimental methods and computational biology methods. For predicting the binding affinity between drug molecules and target proteins, it focuses on introducing three types of methods, i.e. molecular computational simulation, traditional machine learning and deep learning. Moreover, the paper also elaborates the application scenarios, characteristics, pros and cons of each computational biology method. Finally, it discusses the problems and future directions of each prediction method in predicting the binding affinity between drug molecules and target proteins. Taken together, this review aims to provide a reference for developing an efficient model to infer the binding affinity between drug molecules and target proteins.

Key words: binding affinity, drug molecule, target protein, traditional machine learning, deep learning

摘要： 研究药物分子与靶标蛋白结合亲和力有助于了解生物系统和辅助药物开发。随着生物大数据驱动下的计算生物学技术发展，药物分子与靶标蛋白结合亲和力研究策略从传统单一生物医学实验迈向综合计算技术辅助预测，为药物开发提供新技术新方法。鉴于药物分子与靶标蛋白结合亲和力研究的重要性，从传统生物实验方法和计算生物学方法两个维度对其研究进展进行综述，重点介绍了预测药物分子与靶标蛋白结合亲和力的分子计算模拟、传统机器学习和深度学习方法，并阐述了每种计算生物学方法的应用场景、特点、优势和不足。最后，讨论了药物分子与靶标蛋白结合亲和力预测算法存在的问题以及未来方向，旨在为开发高性能药物分子与靶标蛋白结合亲和力预测模型提供参考。

关键词: 结合亲和力, 药物分子, 靶标蛋白, 传统机器学习, 深度学习