深度学习模型在中药毒性预警中的应用和前景

Applications and prospect of deep learning model in early warning of toxicity of traditional Chinese medicine

中药潜在毒性成分的早期筛查是中药新药研发面临的一大难题。在基于“结构预警子-毒性”关系开发的机器学习模型中,使用深度学习算法构建的机器学习模型脱颖而出,有望成为新一代中药毒性预测的杰出工具。本文综述了深度学习模型基于“结构预警子-毒性”关系预测化合物毒性的机制以及深度学习模型在预测药物分子的毒性、预测反应性代谢产物的形成和致毒的生物学靶标中的具体应用,指出中药毒性成分数据匮乏带来的挑战和模型的“黑箱”问题,并提出深度学习模型进一步应用于挖掘中药毒性成分结构和生物学特性中的展望,以期借助人工智能技术解决中药毒性预测这一难题。

Screening potential toxic compounds of traditional Chinese medicine(TCM) is a major challenge to the discovery of new TCM drugs. Among the existing machine learning models,the deep learning(DL) model excels and is expected to become an outstanding tool for safety evaluation of TCM. This article reviews the mechanism of the DL model based on the relationship between "structural alert" and toxicity, and outlines practical applications of the DL model in prediction of drug toxicity, reactive metabolite formation and toxicity-related biological targets. The lack of toxicology data and concerns about the "black box", which currently pose an obstacle to the DL model in TCM, are also discussed.The potential role of the DL model in exploring the structural character and biological property of toxic compounds of TCM is also predicted for the purpose of facilitating closer integration of artificial intelligence and the development of TCM.