谷歌*耶鲁:AI从“工具”到“科学家”,重塑抗癌研发新范式

谷歌DeepMind与耶鲁大学近日联手推出的270亿参数大模型C2S-Scale，正以一场“冷肿瘤”突破，宣告AI制药进入原创假设生成的新纪元。这一合作的价值，远不止于找到一款老药新用，更在于打通了“AI自主提假说—实验室验证—临床转化”的全链路，为全球生物医药研发注入颠覆性动能。

C2S-Scale的核心突破，是将单细胞转录组数据转化为AI可理解的“文本”，让模型像阅读文献一样解析细胞行为。它跳出传统“杀癌”思路，聚焦让免疫系统看见癌细胞，通过虚拟筛选四千余种药物，精准锁定CK2抑制剂Silmitasertib，并提出搭配低剂量干扰素可显著提升抗原呈递的全新假说。耶鲁实验室验证显示，该组合使免疫识别信号暴涨百分之五十，成功将“冷肿瘤”转化为可被攻击的“热肿瘤”。这一发现无文献先例，是AI独立完成的原创科学突破。

传统抗癌药研发需十到十五年、耗资二十到三十亿美元，失败率超百分之九十。而C2S-Scale将数年筛选压缩至数周，且锁定的Silmitasertib已进入临床阶段，老药新用大幅缩短转化周期、降低安全风险。更关键的是，模型实现条件推理，能精准预测药物在不同肿瘤微环境中的效果，让科研资源投向高成功率实验，从根源上减少无效投入。

谷歌与耶鲁将C2S-Scale完全开源，全球实验室均可免费使用。这打破了AI制药的技术壁垒，让资源有限的科研机构也能开展虚拟筛选与假说验证，推动抗癌研究从少数巨头主导走向全球协同创新。模型的开源，更意味着其能力可快速迭代，覆盖更多癌种与疾病领域，形成“AI发现—验证—再发现”的正向循环。

此次合作标志着AI从数据分析工具升级为科学合作者，具备了独立提出原创假说的能力。这不仅重塑抗癌研发逻辑，更将辐射至神经退行性疾病、传染病等领域，开启“AI原生创新”的科研第四纪元。对产业而言，它加速药物研发的去风险化，吸引更多资本投向前沿靶点；对患者而言，意味着更多难治性肿瘤有望迎来低成本、快落地的新疗法。

谷歌与耶鲁的联手，是算力、算法与顶尖生命科学的完美融合。当AI能像科学家一样思考、提出并验证全新假说，人类攻克重大疾病的步伐，必将迎来质的飞跃。

Google × Yale: AI Transforms from “Tool” to “Scientist”, Reshaping the New Paradigm of Anticancer Drug Discovery

Google DeepMind and Yale University have jointly launched C2S-Scale, a large model with 27 billion parameters, announcing a new era of AI-driven pharmaceutical research centered on original hypothesis generation through breakthroughs in “cold tumors”. The value of this collaboration extends far beyond identifying new applications for existing drugs; it establishes a complete workflow of “AI-generated independent hypotheses — laboratory validation — clinical translation”, injecting disruptive momentum into global biomedical research and development.

The core breakthrough of C2S-Scale lies in converting single-cell transcriptomic data into AI-interpretable “text”, enabling the model to analyze cellular behavior much like reading scientific literature. Departing from the traditional cancer-killing approach, it focuses on enabling the immune system to recognize cancer cells. Through virtual screening of more than 4,000 compounds, the model precisely identified Silmitasertib, a CK2 inhibitor, and proposed a novel hypothesis that combining it with low-dose interferon significantly enhances antigen presentation. Validated in Yale’s laboratories, this combination boosted immune recognition signals by 50%, successfully converting “cold tumors” into vulnerable “hot tumors”. This unprecedented discovery represents an original scientific achievement independently generated by AI.

Traditional anticancer drug development takes 10 to 15 years, costs 2 to 3 billion US dollars, and carries a failure rate exceeding 90%. In contrast, C2S-Scale reduces years of screening to weeks. Furthermore, Silmitasertib has already entered clinical trials, and repurposing an approved drug greatly shortens translation cycles and lowers safety risks. Critically, the model supports conditional reasoning, accurately predicting drug efficacy in diverse tumor microenvironments, directing resources toward high-probability experiments and fundamentally reducing inefficient investment.

Google and Yale have fully open-sourced C2S-Scale, making it freely accessible to laboratories worldwide. This breaks technical barriers in AI-driven drug discovery, allowing under-resourced institutions to conduct virtual screening and hypothesis testing. It shifts anticancer research from dominance by a few giants to global collaborative innovation. Open sourcing also accelerates iterative improvement, expanding applications across more cancer types and diseases to form a virtuous cycle of “AI discovery — validation — rediscovery”.

This collaboration marks the evolution of AI from a data analysis tool to a scientific partner capable of independently proposing original hypotheses. It not only reshapes the logic of anticancer research but also extends to neurodegenerative diseases, infectious diseases, and beyond, inaugurating the fourth era of scientific research defined by AI-native innovation. For the industry, it de-risks drug development and attracts greater capital investment in cutting-edge targets. For patients, it brings hope for affordable, rapidly deployed therapies against refractory cancers.

The partnership between Google and Yale represents a perfect integration of computing power, algorithms, and cutting-edge life sciences. As AI gains the ability to think, propose, and validate novel hypotheses like human scientists, humanity’s progress in conquering major diseases will achieve a qualitative leap forward.