深度学习在化学分子逆向合成路线规划中的应用进展
Recent Advances of the Application of Deep Learning for the Retro-synthesis Planning of Chemical Molecules
李帅鑫 1李子昊 2孙婕 2杨旸 3张书宇4
作者信息
- 1. 上海交通大学 致远学院,上海 200240
- 2. 上海交通大学 化学化工学院,上海 200240
- 3. 上海交通大学 电子信息与电气工程学院,上海 200240
- 4. 上海交通大学 致远学院,上海 200240;上海交通大学 化学化工学院,上海 200240
- 折叠
摘要
逆向合成规划是现代有机合成化学中合成路线设计的重要基础.合成化学发展至今,化学家们积累了大量的反应数据.自有机合成大师E.J.Corey将逆合成分析法与计算机结合提出LHASA(logic and heuristics applied to synthetic analysis)起,计算机根据反应数据自主学习并给出逆向合成路线成了化学家的愿景之一.近年来,基于数据驱动的研究范式不断发展,大量深度学习模型被提出并在逆向合成规划中取得了初步的成功,然而该类模型仍然存在高质量数据集稀缺、软硬件结合不佳、领域知识嵌入与发现困难等问题.通过深度学习实现逆向合成路线规划有待深入研究.
Abstract
Retro-synthetic planning stands as a fundamental cornerstone in the design of synthetic routes within modern synthetic organic chemistry.Over the years,chemists have compiled an extensive database of reaction data.Ever since the pioneering work of E.J.Corey,who combined the concept of retro-synthetic analysis with computer algorithms to create LHASA(logic and heuristics applied to synthetic analysis),the vision of comput-ers autonomously learning and proposing retro-synthetic pathways based on reaction data has been a long-standing aspiration among chemists.In recent years,with the evolving data-driven research paradigm,numer-ous deep learning models have been proposed and have achieved preliminary success in retro-synthetic planning.Despite these advancements,the models still confront several challenges,including scarcity of high-quality data-sets,suboptimal integration of software and hardware,and difficulties in embedding and discovering domain-specific knowledge.Therefore,deepening the research to realize retro-synthetic route planning through deep learning remains an imperative endeavor.
关键词
人工智能/有机合成/逆向合成规划/化学信息学Key words
artificial intelligence/organic synthesis/retrosynthesis planning/cheminformatics引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据