Stanford's Virtual Biotech: Multi-Agent AI Framework for Drug Discovery Published in Science
Stanford researchers published a paper in Science on September 17, 2026, describing Virtual Biotech, a multi-agent AI framework that simulates a biotech company's R&D organization for therapeutic discovery. The system comprises four research departments and 11 main agents, with a virtual chief scientific officer coordinating tasks such as target discovery, safety assessment, and clinical development. It uses Claude Sonnet 4.5 and Haiku 4.5 models, and in a data-curation example, deployed 37,075 parallel agent instances to process clinical trial records.
Coverage timeline
机器之心ScienceAI
编辑丨& 一个只存在于想象中的未来企业是什么样的？ 机械、智能体，一切都在有条不紊地自行运转，不受外界干扰，持续朝一个方向努力。这一种名为 Virtual Biotech 的先进 AI 系统，现已于 2026 年 9 月 17 日在《 Science 》的「The Virtual Biotech: A multi-agent AI framework for therapeutic discovery and development」中登场。 这一篇斯坦福大学的研究中，研究团队搭建了一套模拟生物科技公司研发组织的多智能体 AI 系统，让不同的 AI 智能体承担靶点发现、安全评估、治疗方式选择和临床开发等工作，再由一个虚拟首席科学官（CSO）统筹整个过程。 论文链接： https://www.science.org/doi/10.1126/science.aeg6779 Virtual Biotech 药物研发向来是一种吃力不讨好的开发过程，而 AI 的出现为这种繁杂的工作带来了解决的办法。斯坦福大学此次研究的针对的是药物研发中长期存在的一个问题：靶点验证需要同时参考人类遗传学、功能基因组学等多种证据，而这些信息往往分散在不同数据库和专业团队手中。随着数据规模不断扩大，人工完成跨模态整合也越来越困难。 Virtual Biotech 因此按照一家真实药企的方式组织 AI。系统有四个研究部门和 11 个主要智能体，底层语言模型采用 Claude Sonnet 4.5，负责战略简报和科学审查的两个智能体则使用 Claude Haiku 4.5。它们通过 MCP 服务器连接生物医学数据库、代码执行环境和分析工具，让语言模型能够直接调用数据和计算资源。 图 1：Virtual Biotech 智能体、工具及编排流程概述。 数据整理 这条路流程最醒目的数据来自于临床实验数据整理。比如，研究人员希望把 Open Targets 中的临床试验补充上更加完整的终点结果和不良事件信息。 Virtual Biotech 先让临床试验智能体设计数据提取规则，然后针对不同的临床试验编号，同时派出 37,075 个临床试验智能体实例进行并行处理。每个智能体负责一项试验，先过一遍 ClinicalTrials.gov，再根据需要检索 PubMed 以及新闻稿和监管公告，并记录每条信息的来源。 图
Related stories
Tsinghua introduces SRHarness and MechBench for agentic symbolic regression and mechanism discovery
Tsinghua University researchers introduced SRHarness, a runtime framework for agentic symbolic regression, and MechBench, a benchmark for mechanism discovery, aiming to improve AI's ability to derive formulas from data and understand underlying physical mechanisms. Experiments show SRHarness achieves 72.97% symbol recovery with DeepSeek-v4-flash-0731, far exceeding Codex's 20.72%, while MechBench reveals that even GPT-5.6-sol-powered Codex recovers mechanisms at only 13.75% on the core set, highlighting a significant gap between pattern discovery and mechanism understanding.
1 reports · 1 sources
Federated learning across 5 pharma firms improves OpenFold3 drug-interaction predictions
A Nature paper (Sept 14, 2026) reports that researchers used federated learning across proprietary data from five pharmaceutical companies—AbbVie, Astex Pharmaceuticals, Bristol Myers Squibb, Johnson & Johnson, and Takeda—to improve the OpenFold3 protein model. Adding over 20,000 proprietary protein–small-molecule structures, the model's protein–drug interaction predictions outperformed models trained only on public data or on single-company data. The work addresses the scarcity of experimentally determined drug-like interaction structures in public databases like PDB.
VideoGen-Agent: 8B multimodal agent learns tool use for video generation via RL
Researchers released VideoGen-Agent, an 8B multimodal agent built on Qwen3-VL-8B-Instruct that learns to call retrieval, generation, and verification tools for video generation through supervised fine-tuning and multi-task reinforcement learning. On the VABench benchmark with 600 prompts covering six capabilities, its Toolset 1 configuration scored 75.6, up 19.1 from the base generator's 56.5, and reached 86.1 when generation tools were upgraded without changing agent parameters. In human comparisons, evaluators preferred the upgraded configuration's videos over the strongest single-model baseline in 84.3% of judgments.
Google DeepMind, Meta, Isomorphic Labs invest $300M in Biohub virtual cell project
Google DeepMind, Meta, and Isomorphic Labs are jointly investing $300 million into Biohub, the nonprofit biomedical research organization founded by Mark Zuckerberg and Priscilla Chan. The funding supports a $1.8 billion initiative to build AI datasets for a 'virtual cell' that could enable digital biological simulations to help prevent and manage diseases.
1 reports · 1 sources
News updates every hour. Subscribe in your RSS reader to get new stories automatically.
Subscribe