基于机器学习的药物-靶标相互作用预测*

近年来,随着计算机硬件、软件工具和数据丰度的不断突破,以机器学习为代表的人工智能技术在生物、基础医学和药学等领域的应用不断拓展和融合,极大地推动了这些领域的发展,尤其是药物研发领域的变革。其中,药物-靶标相互作用(drug-target interactions, DTI)的识别是药物研发领域中的重要难题和人工智能技术交叉融合的热门方向,研究人员在DTI预测方面做了大量的工作,构建了许多重要的数据库,开发或拓展了各类机器学习算法和工具软件。对基于机器学习的DTI预测的基本流程进行了介绍,并对利用机器学习预测DTI的研究进行了回顾,同时对不同的机器学习方法运用于DTI预测的优缺点进行了简单总结,以期对开发更加有效的预测算法和DTI预测的发展提供帮助。

Research Progress of Drug Target Interaction Prediction Based on Machine Learning

In recent years, with the continuous breakthrough of computer hardware capability, software efficiency and data abundance, the applications of artificial intelligence technology represented by machine learning have been continuously expanded and integrated, which has greatly promoted the development in fields of biology, medicine, pharmacy, and especially drug R&D. Among those technology advancements, the identification of drug-target interactions (DTI) is an important problem in the field of drug R&D and a popular research direction for the cross-integration of artificial intelligence technology. As the source of innovative drug development, drug-target interaction prediction can provide high-probability potential drug targets for biological experiments, thereby increasing the rate of lead compound discovery, increasing the success rate of late-stage drug development and shortening the total development cycle. Researchers have already done a lot of work in constructing the prediction methods of drug-target interactions by building databases, developing software and establishing machine learning algorithms. In most works, data are transformed into feature vectors or similarities, and then suitable machine learning methods are employed to build predictive models. This paper introduces the basic process and reviews the research progress of drug-target interaction prediction based on machine learning. In addition, the advantages and disadvantages of existing prediction methods are briefly summarized in order to facilitate the development of more efficient prediction algorithms and drug-target interaction prediction methods.