来自杂志 Zenodo 的文献。
当前共找到 1 篇文献分享。
1.
小年
(2026-03-31 22:08):
#paper arXiv:2603.12457(预印本)，The Single-Model Illusion in AI-Driven Drug Discovery: Introducing a Systems-Level Multi-Model Framework for Translational Discovery
研究团队针对当前AI驱动药物研发普遍依赖单一模型带来的预测偏差、泛化能力弱、临床转化成功率低等“单模型错觉”问题，提出了一套系统级多模型整合框架。该研究通过对比分析单一预测模型在分子设计、靶点结合、药代动力学及毒性评估中的局限性，揭示了过度依赖单一会导致研究结果与临床实际脱节。在此基础上构建的多模型协同体系，整合了靶点建模、分子生成、理化性质预测、细胞与动物水平验证等多层级计算模型，实现从分子筛选到转化研究的全流程交叉验证与决策优化。实际测试表明，该框架能有效降低假阳性与模型误导，提升候选药物的可靠性与可转化性，为突破现有AI制药瓶颈、建立更稳健的转化式药物发现体系提供了新的研究范式。
Zenodo,
2026/3/26.
DOI: 10.5281/zenodo.19240171
Abstract:
Recent advances in AI-driven drug discovery have led to widespread narratives suggesting that a single model or platform can generate viable therapeutic candidates and, when combined with automated laboratory systems, …
>>>
Recent advances in AI-driven drug discovery have led to widespread narratives suggesting that a single model or platform can generate viable therapeutic candidates and, when combined with automated laboratory systems, rapidly progress to clinical development. These narratives often imply that AI-driven design coupled with robotic execution can substantially compress the path to Phase I trials and accelerate the treatment of complex diseases within a few years. However, practical implementation reveals a significant gap between model-level performance and end-to-end drug development success.
<<<
翻译