可解释性
药物发现
计算机科学
决策支持系统
优先次序
人工智能
数据科学
可信赖性
临床决策支持系统
任务(项目管理)
机器学习
领域(数学分析)
风险分析(工程)
药物重新定位
决策树
分子描述符
可测试性
领域知识
知识管理
药品
适用范围
管理科学
知识抽取
虚拟筛选
标识
DOI:10.1016/j.drudis.2026.104679
摘要
• Physics-inspired explainable artificial intelligence links prediction to mechanistic insight, improves plausibility and testability of molecular hypotheses and enables decision support in lead optimization and toxicity. • Decision-centric metrics assess practical utility. • Molecular intelligence frames decision-driven workflows. Explainable artificial intelligence (XAI) is increasingly recognized as essential for trustworthy drug discovery, yet many approaches remain post hoc and correlational. This review examines physics-inspired XAI as a framework for improving mechanistic interpretability and decision support at the molecular design stage. We highlight its strengths in modeling molecular recognition and local structure–property relations, while emphasizing that efficacy and toxicity emerge from biological and pharmacological processes beyond molecular physics. Current evidence suggests that explainability improves prioritization and experimental efficiency, although its impact on overall R&D outcomes remains limited. We propose ‘molecular intelligence’ as a decision-centric framework integrating prediction, validation and domain knowledge.
科研通智能强力驱动
Strongly Powered by AbleSci AI