arXiv cs.AI· Jiaxing Sun, Xuli Shen, Jiang Wu, Ruiyi Ding, Ruikang Zhang, Xiao An, Xibao Li, Yubin Wang, Xingjian Wei, Yinfan Wang, Yu Liu, Lingli Ge, Xi Xue, Xiaojian Wang, Lijun Wu, Chang-Yu Hsieh, Yuqiang Li, Bowen Zhou, Dahua Lin, Conghui He·· 14 小时前AI 评分29
反应先例让逆合成预测突破训练覆盖范围
Reaction precedents extend retrosynthetic prediction beyond training coverage
AI 导读
研究用 271 万条反应数据限制训练与检索覆盖后发现，仅扩展检索语料、固定模型参数，被排除训练的转化类型 Top-1 准确率从 13.0% 升至 58.0%，时间留出测试集从 4.9% 升至 23.1%。仅用 2017 年前数据训练的模型经检索扩展后，在两个测试集上追平或超过用 2025 年前数据训练的无检索模型。14 个通用模型在无需任务微调下均获提升，最强模型超过 8 个专用模型。
正文 · 原文
Authors:Jiaxing Sun, Xuli Shen, Jiang Wu, Ruiyi Ding, Ruikang Zhang, Xiao An, Xibao Li, Yubin Wang, Xingjian Wei, Yinfan Wang, Yu Liu, Lingli Ge, Xi Xue, Xiaojian Wang, Lijun Wu, Chang-Yu Hsieh, Yuqiang Li, Bowen Zhou, Dahua Lin, Conghui He
Abstract:Reaction precedents guide retrosynthesis, but how far they can compensate for gaps in task-specific training coverage remains unclear. Using 2.71 million reactions, we restricted training and retrieval coverage by transformation or time, then expanded only the retrieval corpus with model parameters fixed. For transformations excluded from training, mean Top-1 accuracy increased from 13.0% to 58.0% on a historical test set and from 4.9% to 23.1% on a temporally held-out test set. For temporal coverage gaps, retrieval expansion enabled a model trained on records before 2017 to match or exceed a non-retrieval model trained on records before 2025 on both test sets. Without task-specific fine-tuning, precedents improved all 14 representative general-purpose models on both test sets and USPTO-50K; the strongest with precedents surpassed all eight representative specialized models. These findings show that inference-time reaction precedents can extend retrosynthetic prediction beyond task-specific training coverage without parameter updates.
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.40249 [cs.AI] |
| (or arXiv:2610.40249v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.40249 arXiv-issued DOI via DataCite |
Submission history
From: Jiang Wu [view email]
[v1]
Thu, 29 Oct 2026 15:19:54 UTC (20,552 KB)
来源：arXiv cs.AI · arxiv.org