---
格式版本: 2
标题: "Monroe: A Molecular Foundation Model for In-Context Probabilistic Inference"
原文链接: "https://arxiv.org/abs/2608.18982"
发布日期: "2026-08-19"
发布时间校准状态: "found"
发布时间需复核: "否"
发布时间来源: "rule:local:strict_original_body"
发布时间证据: "**\\[v1\\]** Wed, 19 Aug 2026 14:48:19 UTC (1,102 KB)"
发布时间校准原因: "规则确认唯一严格发布时间，来源 local:strict_original_body"
发布时间校准置信度: "high"
发布时间候选数量: 18
发布时间严格候选数量: 6
发布时间原页读取状态: "source template page reused from URL open"
发布时间未找到原因: ""
发布时间校准时间: "2026-08-20T15:35:04+08:00"
发布时间仲裁状态: "skipped"
发布时间仲裁尝试次数: 0
发布时间仲裁耗时毫秒: 0
发现时间: "2026-08-20T15:26:21+08:00"
入库时间: "2026-08-20T07:35:04.318Z"
来源平台: "arXiv 学术论文搜索"
搜索渠道: "source_template"
搜索词: "https://arxiv.org/search/?query=performance&searchtype=all"
匹配关键词:
  - "performance"
相关厂家:
  []
相关专家:
  []
内容类型: "网页"
抓取工具: "Free Fetch + Defuddle"
清洗工具: "Defuddle Markdown + Defuddle/Readability 正文提取"
原始附件:
  []
AI优质: "否"
AI打分: 5
AI分档: "非优质"
AI质检状态: "不通过"
AI打分理由: "该资料为arXiv分子基础模型论文，主题涉及生物医药机器学习，与超节点、AI Rack、机柜级AI基础设施、供电散热互连等完全无关，属于明显无关内容。"
AI质检模型: "ali-deepseek-v4-flash"
AI质检时间: "2026-08-20T15:37:50+08:00"
AI主题相关性: 0
AI来源权威性: 5
AI新颖性: 0
AI技术细节: 0
AI商业部署信号: 0
AI完整性: 5
AI摘要: "Monroe 是一个新的分子基础模型，在 8100 万分子的量子化学数据集上预训练，并通过改进立体化学图表示、训练损失和引入 TabPFN 进行上下文概率预测。在 Polaris 基准上它匹配或超过现有模型，在活性悬崖基准上显著提升；"
AI摘要模型: "ali-deepseek-v4-flash"
AI摘要时间: "2026-09-07T03:19:53.719Z"
采集批次: "2026年8月20日14点19分32秒"
采集批次ID: "20260820-141932-079"
去重键: "https://arxiv.org/abs/2608.18982"
---

## Computer Science > Machine Learning

## Title:Monroe: A Molecular Foundation Model for In-Context Probabilistic Inference

Authors:[Blazej Banaszewski](https://arxiv.org/search/cs?searchtype=author&query=Banaszewski,+B), [Andrew W. Fitzgibbon](https://arxiv.org/search/cs?searchtype=author&query=Fitzgibbon,+A+W)

[View PDF](https://arxiv.org/pdf/2608.18982) [HTML (experimental)](https://arxiv.org/html/2608.18982v1)

> Abstract:Bioassay activity prediction is often data-limited because drug-discovery datasets rely on time-consuming and expensive wet-lab experiments for data generation and evaluation. This challenge has inspired recent research into molecular foundation models (MFMs), which aim to encode general-purpose chemical knowledge into molecular representations that generalize well in data-constrained scenarios. This paper presents Monroe, a new MFM with several innovations over the existing state of the art: increased scale allowing pre-training on over 81 million molecules from the PM6 quantum chemistry dataset; improved graph representation of stereochemistry; improved training losses including conformer denoising and embedding decorrelation; improved multi-task learning; and the use of a prior-data-fitted model (TabPFN) for downstream in-context prediction. Our evaluations use a principled pairwise comparison framework that measures statistically significant performance differences. Across established Polaris benchmarks, Monroe matches or exceeds existing MFMs, while on activity cliff benchmarks, designed to assess utility for molecular discovery, it achieves significant improvements over prior methods. Finally, ablation and transfer experiments show that PFN-based downstream predictors also substantially improve two leading existing models, MiniMol and CheMeleon, yielding new state-of-the-art variants we call MiniMol\_PFN and CheMeleon\_PFN, suggesting that our downstream adaptation strategy generalizes beyond Monroe. Source code is at [this http URL](http://github.com/blazejba/monroe).

| Comments: |  |
| --- | --- |
| Subjects: | Machine Learning (cs.LG); Biomolecules (q-bio.BM); Quantitative Methods (q-bio.QM) |
| Cite as: | [arXiv:2608.18982](https://arxiv.org/abs/2608.18982) \[cs.LG\] |
|  | (or [arXiv:2608.18982v1](https://arxiv.org/abs/2608.18982v1) \[cs.LG\] for this version) |
|  | [https://doi.org/10.48550/arXiv.2608.18982](https://doi.org/10.48550/arXiv.2608.18982) |

## Submission history

From: Blazej Banaszewski \[[view email](https://arxiv.org/show-email/d2df3374/2608.18982)\]  
**\[v1\]** Wed, 19 Aug 2026 14:48:19 UTC (1,102 KB)

[Which authors of this paper are endorsers?](https://arxiv.org/auth/show-endorsers/2608.18982) | Disable MathJax ([What is MathJax?](https://info.arxiv.org/help/mathjax.html))