Yang Liu

Yu Gu、Eric Topol 和 Hoifung Poon 团队用压力测试说明，健康 AI 的 benchmark 高分还不能直接等同于临床可用性。

The Yu Gu, Eric Topol and Hoifung Poon teams show that high health AI benchmark scores are not yet the same as clinical readiness.

这项 Nature 研究首次用 ABE8e 碱基编辑在人类胚胎中功能扰动 NANOG，显示 NANOG 对上胚层指定必不可少，同时揭示人类 primitive endoderm 命运与小鼠不同。

This Nature study uses ABE8e base editing to functionally perturb NANOG in human embryos, showing that NANOG is essential for epiblast specification while revealing a human-specific retention of primitive endoderm fate.

CVSP-AIE 把 KarmaDock、CarsiDock 和 RTMScore 封装成可运行的 AI 虚拟筛选流程，真正加速的是候选排序；真实命中仍要靠湿实验闭环。

CVSP-AIE packages KarmaDock, CarsiDock and RTMScore into an executable AI virtual-screening workflow; the real acceleration is candidate ranking, while real hit proof still needs wet-lab closure.

这篇 Blood 研究把铁过载对 HSC 的伤害定位到线粒体铁、mtROS 和 OXPHOS 下降，并显示 DFO 或 MitoQ 可部分恢复 HSC 静息与自我更新。

This Blood study maps iron-overload injury in HSCs to mitochondrial iron, mtROS, reduced OXPHOS and glycolytic rewiring, with partial rescue by DFO or MitoQ.

OpenIO 是一篇 AI-native immunotherapy 路线图，价值在于把数据、模型、闭环实验和数字免疫孪生放到同一张地图上；但它还不是平台已被验证的证据。

OpenIO is an AI-native immunotherapy roadmap: valuable as a map linking data, models, closed-loop experiments and digital immune twins, but not yet evidence that the platform is validated.

儿童 exa-cel 证明编辑细胞可以很干净地起效, 也暴露出血液基因治疗真正难以规模化的部分: 不是编辑, 而是为了让编辑细胞住进骨髓所付出的清髓代价。

Pediatric exa-cel shows that edited cells can work cleanly, while exposing the hardest scaling problem in blood gene therapy: the price paid to open marrow niches.

礼来的 AI-bio 大单不是现金胜利，而是用 milestone 把临床前平台期权提前定价。

Lilly's AI-bio deals are not proof of clinical victory; they are milestone-heavy options on preclinical platforms.

AI 在科学里快在能被便宜验证的地方,慢在验证昂贵又主观的地方;而最汹涌的资本,正押在验证的前面、跑向终点。