---
格式版本: 2
标题: "Leveraging generative hallucination and biophysics-informed modeling for unified biomolecular sequence-structure co-design"
原文链接: "https://arxiv.org/abs/2608.17381"
发布日期: "2026-08-18"
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\\]** Tue, 18 Aug 2026 05:16:41 UTC (542 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: "2026-08-20T14:45:28+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: "2026-08-20T14:38:42+08:00"
入库时间: "2026-08-20T06:45:28.724Z"
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=Oracle&searchtype=all"
匹配关键词:
  - "AI"
相关厂家:
  - "Oracle"
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: 0
AI分档: "非优质"
AI质检状态: "不通过"
AI打分理由: "该资料为arXiv生物分子序列-结构共设计论文，内容完全与超节点、AI Rack、机柜级AI基础设施无关，仅因搜索词含Oracle而命中，实为无关资料。"
AI质检模型: "ali-deepseek-v4-flash"
AI质检时间: "2026-08-20T14:48:43+08:00"
AI主题相关性: 0
AI来源权威性: 0
AI新颖性: 0
AI技术细节: 0
AI商业部署信号: 0
AI完整性: 0
AI摘要: "研究者提出MCTH框架，融合生成式幻觉与生物物理约束建模，实现多模态生物分子序列与结构共设计；"
AI摘要模型: "ali-deepseek-v4-flash"
AI摘要时间: "2026-09-07T03:21:22.389Z"
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/2608.17381"
---

## Quantitative Biology > Quantitative Methods

## Title:Leveraging generative hallucination and biophysics-informed modeling for unified biomolecular sequence-structure co-design

Authors:[Xuefeng Liu](https://arxiv.org/search/q-bio?searchtype=author&query=Liu,+X), [Mingxuan Cao](https://arxiv.org/search/q-bio?searchtype=author&query=Cao,+M), [Xiao Luo](https://arxiv.org/search/q-bio?searchtype=author&query=Luo,+X), [Songhao Jiang](https://arxiv.org/search/q-bio?searchtype=author&query=Jiang,+S), [Tobin Sosnick](https://arxiv.org/search/q-bio?searchtype=author&query=Sosnick,+T), [Jinbo Xu](https://arxiv.org/search/q-bio?searchtype=author&query=Xu,+J), [Louis Maher](https://arxiv.org/search/q-bio?searchtype=author&query=Maher,+L), [Rick Stevens](https://arxiv.org/search/q-bio?searchtype=author&query=Stevens,+R)

[View PDF](https://arxiv.org/pdf/2608.17381) [HTML (experimental)](https://arxiv.org/html/2608.17381v1)

> Abstract:Biomolecular design underpins applications from molecular recognition to therapeutics and synthetic biology, yet de novo interaction design remains challenging-especially for DNA/RNA, underexplored non-protein modalities with scarce, heterogeneous complex data and sharper geometric and chemical constraints. We introduce MCTH (Monte Carlo Tree Hallucination), an inference-only framework that casts all-atom sequence-structure co-design as uncertainty-aware planning over hallucinated states from pretrained folding and inverse-folding models, with optional biophysical control within the same decision loop. MCTH treats these models as frozen black-box operators and uses Monte Carlo Tree Search to allocate a fixed inference budget across competing design trajectories, incorporating model confidence and uncertainty, as well as cross-expert consensus/disagreement when multiple predictors are available. Across protein-RNA, protein-DNA, protein-protein, and protein-ligand design, matched-budget experiments show that adaptive search improves over simpler sampling and cycling strategies, while held-out AlphaFold3 and Chai-1 evaluations demonstrate transfer beyond the search-time oracle. MCTH provides a shared planning layer across modalities while allowing task-specific folding, inverse-folding, and biophysical modules, requiring no fine-tuning or backpropagation through component models.

| Subjects: | Quantitative Methods (q-bio.QM); Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Biomolecules (q-bio.BM) |
| --- | --- |
| Cite as: | [arXiv:2608.17381](https://arxiv.org/abs/2608.17381) \[q-bio.QM\] |
|  | (or [arXiv:2608.17381v1](https://arxiv.org/abs/2608.17381v1) \[q-bio.QM\] for this version) |
|  | [https://doi.org/10.48550/arXiv.2608.17381](https://doi.org/10.48550/arXiv.2608.17381) |

## Submission history

From: Xuefeng Liu \[[view email](https://arxiv.org/show-email/7c3be09a/2608.17381)\]  
**\[v1\]** Tue, 18 Aug 2026 05:16:41 UTC (542 KB)

[Which authors of this paper are endorsers?](https://arxiv.org/auth/show-endorsers/2608.17381) | Disable MathJax ([What is MathJax?](https://info.arxiv.org/help/mathjax.html))