arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2607.01800cs.LGcs.CL
收藏

LLMs 在分子领域真的能泛化吗?基于扰动的分析

Do LLMs Truly Generalize in the Molecular Domain? A Perturbation-Based Analysis

The Hong Kong Polytechnic University(香港理工大学);National University of Singapore(新加坡国立大学);Shanghai AI Lab(上海人工智能实验室);Fudan University(复旦大学)

Jiatong Li, Weida Wang, Changmeng Zheng, Shufei Zhang, Yatao Bian, Xiao-yong Wei, Qing Li

AI总结:

提出分子扰动框架,通过控制图编辑距离生成结构变体,揭示分子LLMs对微小结构变化敏感,局部信任区域狭窄;上下文调优可部分缓解脆弱性。

摘要:

大型语言模型(LLMs)最近在分子发现中展现出潜力,但其基于离散序列令牌的概率性质与化学空间的刚性拓扑约束之间仍存在差距。这引出一个问题:分子LLMs能否在基于序列表示的局部邻域之外泛化?为系统研究此问题,我们引入分子扰动框架,在受控图编辑距离(GED)下生成训练分子的语法有效结构变体,以探测分子LLMs的流形规律性。分析表明,即使单个编辑也可能导致常见分子任务性能显著下降,揭示了狭窄的局部信任区域和对结构变化的脆弱敏感性。由于相似分子往往表现出相似性质,上下文调优(ICT)通过将预测锚定在结构相似的分子上,提供了一种自然的缓解方式。我们的实验还检验了ICT在受控结构扰动下是否能赋予鲁棒性,结果表明它能在一定程度上扩展局部信任区域,并为稳定分子LLMs对抗结构变化提供了有前景的方向。

英文摘要:

Large Language Models (LLMs) have recently shown promise in molecular discovery, yet a gap remains between their probabilistic nature over discrete sequential tokens and the rigid topological constraints of chemical space. This raises the question of whether molecular LLMs can generalize beyond the local neighborhoods induced by their sequence-based representations. To systematically investigate this question, we introduce a Molecular Perturbation framework that generates syntax-valid structural variants of training molecules under controlled Graph Edit Distance (GED) to probe the manifold regularity of molecular LLMs. Our analysis shows that even a single edit can cause substantial performance drops on common molecular tasks, revealing a narrow local trust region and fragile sensitivity to structural changes. Since similar molecules tend to exhibit similar properties, In-Context Tuning (ICT), which anchors predictions on structurally similar molecules, offers a natural way to mitigate such fragility. Our experiments also examine whether ICT confers robustness under controlled structural perturbations, and the results suggest that it can partially expand the local trust region and offer a promising direction for stabilizing molecular LLMs against structural variation.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑