OpenAI 联手 PerturbAI 让近 2000 个 AI 智能体一起搞科研 加速新药研发OpenAI Partners with PerturbAI: Nearly 2,000 AI Agents Team Up to Accelerate Drug Discovery


AI free token/AI免费积分
OpenAI 联手 PerturbAI 让近 2000 个 AI 智能体一起搞科研 加速新药研发
感谢IT之家网友 补药吖 的线索投递!

北京时间 10 月 21 日,生物工艺公司 PerturbAI 宣布与 OpenAI 合作的最新进展。双方联合斯克里普斯研究所(Scripps Research)开发了一套 AI 推理框架,借助近 2000 个 AI 智能体并行工作,协助科学家分析复杂生物实验数据、检索文献,并提出可供后续实验检验的假说。

据 PerturbAI 介绍,这套平台在分析其全脑基因组图谱时,用 32 分钟完成数据分析与研究发现整合。公司估计,如果由人类按串行方式开展分析和文献审阅,相关工作可能需要数月至一年。这一时间比较来自公司估算,并非与人类团队开展的同期对照测试。

在具体流程中,AI 智能体分别分析细胞反应、搜索科学文献,将实验结果与已有生物学知识联系起来,再汇总形成“推理图谱”,串联基因、细胞反应、既有研究及可能的生物学机制,为科学家给出进一步调查的线索。

研究团队还邀请生物学专家,从新颖性、证据兼容程度、后续研究优先级和置信度等维度评估报告。在与单个 AI 智能体生成报告的盲评比较中,专家有 75% 的时候更偏好多智能体平台给出的解释。这一结果反映的是专家对报告的偏好,不能直接理解为研究结论的准确率。

此次分析使用的数据来自 PerturbAI 今年 3 月公布的全脑 CRISPR 图谱,涵盖约 800 万个细胞。公司称,该图谱在数月内生成,当下下载量已超过 4 万次。此后,PerturbAI 还构建了覆盖代谢疾病和衰老研究的专有图谱。

双方的合作已持续约一年,接下来将利用 PerturbAI 在活体平台中生成的因果生物学数据,联合开展新药靶点发现。相关成果已借助联合预印本和博客公布,当下披露的结果主要涉及数据解读及研究假说生成,所提出线索仍需后续实验验证。

本文来源:https://www.ithome.com/1/016/134.htm
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。

On October 21, Beijing time, bioprocess company PerturbAI announced the latest progress of its cooperation with OpenAI. Together with Scripps Research, the two companies have developed an AI reasoning framework that uses nearly 2,000 AI agents to work in parallel to assist scientists in analyzing data from complex biological experiments, retrieving literature, and proposing hypotheses that can be tested in subsequent experiments.

According to PerturbAI, this platform analyzes its whole brain genome map in 32 minutes to complete data analysis and research findings integration. The company estimates that the work could take months to a year if analyses and literature reviews were carried out by humans in a serial fashion. This time comparison comes from company estimates and is not a contemporaneous controlled test with the human team.

In the specific process, the AI agent separately analyzes the cellular response, searches the scientific literature, links the experimental results with the existing biological knowledge, and then summarizes and forms a "reasoning map", connecting genes, cellular responses, existing research, and possible biological mechanisms to give scientists clues for further investigation.

The research team also invited biology experts to evaluate reports from the dimensions of novelty, evidence compatibility, follow-up research priorities, and confidence. In the blind evaluation comparison with the report generated by a single AI agent, experts prefer the explanation given by the multi-agent platform 75% of the time. This result reflects the experts' preference for reporting and cannot be directly understood as the accuracy of the study conclusions.

The analysis used data from PerturbAI's whole brain CRISPR atlas released in March this year, covering about 8 million cells. The Atlas was generated in a matter of months and has now been downloaded more than 40,000 times, the company said. Since then, PerturbAI has also constructed a proprietary atlas covering metabolic diseases and aging research.

The collaboration has been ongoing for about a year, and the next step will be to use the causal biology data generated by PerturbAI in the live platform to jointly carry out new drug target discovery. Relevant results have been published by means of joint preprint and blog. At present, the results disclosed mainly involve data interpretation and research hypothesis generation. The clues proposed still need to be verified by follow-up experiments.

This article is from: https://www.ithome.com/1/016/134.htm For copyright issues, please write to: 13951431913#139.com or submit a review on the right.

OpenAI Partners with PerturbAI: Nearly 2,000 AI Agents Team Up to Accelerate Drug Discovery

暂无评论,快来发表第一条评论吧!No comments yet. Be the first!

📮 需求咨询📮 Inquiry