Journal

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.

这篇 Cell 论文把 AI 放进 CAR-T 靶点发现链路，真正加速的是候选靶点排序；昂贵证明仍然落在安全性、人体疗效和临床闭环。

This Cell paper places AI inside CAR T target discovery: the real acceleration is candidate triage, while the expensive proof remains safety, human efficacy and clinical validation.

ERA 把科学软件写作转成可自动评分的代码搜索问题，显示 AI 在验证便宜的计算任务上能快速试错，但不能把 leaderboard 分数等同完整科学发现。

ERA turns scientific software development into an automatically scored code-search loop, showing that AI moves fastest when validation is cheap and objective.

这篇 Review 把 AI 靶点发现拆成数据源、模型、靶点评估和验证阶梯；它的价值在于给出领域地图，也提醒靶点验证仍然卡在实验与临床闭环。

This Review maps AI target discovery across data sources, model families, assessment criteria and validation levels; its value is the field map, and its caveat is that target validation still depends on experimental and clinical closure.

这篇 Perspective 提出 MRAI 框架，把大推理模型定位为医疗里的可审计推理伙伴；真正卡点不是模型会不会推理，而是推理能不能被低成本验证、治理和追责。

This Perspective proposes MRAI as an auditable medical-reasoning partner and makes clear that the hard part is not just model reasoning, but low-cost verification, governance and accountability.

Germinal 把 AI 放在抗体设计价值链的设计段，用结构预测和抗体语言模型把候选压缩到每个靶点几十个，再用湿实验闭环验证。

Germinal puts AI into the design step of antibody discovery, using structure prediction and antibody language modeling to shrink experimental validation to dozens of candidates per target.

这篇 Cell Stem Cell 研究显示，CBE 可在 FHL3/Jinx 小鼠模型中恢复免疫功能，但基因毒性图谱同时强烈依赖编辑平台、细胞类型和时间。

This Cell Stem Cell study shows that CBE can restore immune function in an FHL3/Jinx mouse model, while genotoxicity depends strongly on editing platform, cell type, and time.

这篇 Blood 研究在 4 只非人灵长类中显示，humanized CD45 antibody 搭载 alpha emitter astatine-211 可支持 multiplex-edited autologous HSPC 的快速恢复、长期多谱系植入和高频单细胞编辑读出。

This Blood study shows in four nonhuman primates that a humanized CD45 antibody carrying astatine-211 can support rapid recovery, durable multilineage engraftment, and high-frequency single-cell editing readouts after multiplex-edited autologous HSPC infusion.

这篇 NEJM III 期随机双盲试验显示，单次 50 mg lonvo-z 肝靶向在体 CRISPR 编辑 KLKB1，可在第 5-28 周把 HAE 月发作率相对降低 87%，但长期编辑安全性仍需随访。

This NEJM phase 3 randomized trial shows that one 50 mg liver-targeted in vivo CRISPR dose of lonvo-z editing KLKB1 reduced HAE monthly attack rate by 87% from weeks 5 to 28, while long-term editing safety still needs follow-up.

这篇儿科 exa-cel 中期数据最重要的信息不是 CRISPR 又一次有效，而是疗效信号已经足够清楚，真正限制儿童治愈性基因治疗的是 busulfan 清髓带来的代价。

The most important message from this pediatric exa-cel interim analysis is not that CRISPR works again, but that the efficacy signal is clear while the limiting risk now sits in busulfan conditioning.

这篇 Nature Microbiology Brief Communication 用 APEX 1.1 从 2,897 个 prion 相关蛋白中筛出 1,179 个候选 prionins，并证明其中一批具有抑菌、膜扰动和小鼠局部感染模型活性。

This Nature Microbiology Brief Communication uses APEX 1.1 to mine 1,179 candidate prionins from 2,897 prion-related proteins, then validates antibacterial, membrane-perturbing and mouse infection-model activity in a subset.

这篇 Nature 论文整合超过 11,000 个跨组织、跨物种转录组，建立衰老和预期死亡风险时钟，并把死亡相关信号拆解为炎症、线粒体、染色质等可解释模块。

This Nature paper integrates more than 11,000 cross-tissue, cross-species transcriptomes to build ageing and expected-mortality clocks, then decomposes mortality-linked signals into interpretable inflammatory, mitochondrial, chromatin and other modules.

这篇 Nature Nanotechnology 论文系统优化全 RNA PE-LNP，在小鼠肝脏实现单剂 2 mg/kg、49% Pcsk9 prime editing，并在 PAH R408W PKU 小鼠中达到疾病救援水平。

This Nature Nanotechnology paper systematically optimizes all-RNA PE-LNPs, achieving 49% Pcsk9 prime editing in mouse liver with a single 2 mg/kg dose and disease-rescue-level editing in PAH R408W PKU mice.

这篇 NEJM Brief Report 报道一名 MPSI/Hurler 患儿在枕大池内 AAV9-IDUA 给药 4 年后发生 PLAG1 驱动的神经上皮肿瘤，肿瘤中检出重排 AAV 调控元件整合和 AAV-PLAG1 融合转录本。

This NEJM Brief Report describes a PLAG1-driven neuroepithelial tumor 4 years after intracisternal magna AAV9-IDUA delivery in a child with MPSI/Hurler syndrome, with rearranged AAV regulatory elements integrated into PLAG1 and an AAV-PLAG1 fusion transcript.

这篇 Nature Biotechnology 论文用 prime editor 建立 PRESENT 高通量筛选，显示人类细胞中多数同义突变接近中性，但少数可通过剪接、RNA 折叠和翻译影响细胞适应度。

This Nature Biotechnology paper uses prime editing to build PRESENT, showing that most synonymous variants are near-neutral in human cells while a small subset affects fitness through splicing, RNA folding and translation.

这篇 Nature Biomedical Engineering 论文提出 PRIME-In，用 prime-edited donor flap 与 genomic nicking 实现 DSB-free 大片段靶向整合，并把它推进到原代人 T 细胞和 CD19 CAR-T 制备。

This Nature Biomedical Engineering paper introduces PRIME-In, a DSB-free large-DNA targeted integration strategy that couples a prime-edited donor flap with genomic nicking and advances it into primary human T-cell and CD19 CAR-T engineering.

这篇 Nature Medicine phase 1 论文把工程化自体 HSPC 用作持续生成髓系细胞的 IFNα 局部递送平台，在新诊断 MGMT 未甲基化 GBM 中显示安全性、可行性和肿瘤微环境重塑信号。

This Nature Medicine phase 1 paper uses engineered autologous HSPCs as a durable myeloid-cell source for local IFNα delivery in newly diagnosed MGMT-unmethylated GBM, showing safety, feasibility and tumor-microenvironment reprogramming signals.

这篇 bioRxiv 预印本提出 Shred-seq，用 CRISPR-Cas3 大片段 deletion 加 phage-polymerase breakpoint genotyping，在人类基因组 14% 区间上直接测量 coding 与 noncoding essentiality。

This bioRxiv preprint introduces Shred-seq, combining CRISPR-Cas3 large deletions with phage-polymerase breakpoint genotyping to directly measure coding and noncoding essentiality across 14% of the human genome.

这篇 Review 把疫苗之后的 mRNA 药物化路线整理成一张转化地图：真正的瓶颈不再只是 mRNA 设计，而是递送、免疫静默、表达时长和重复给药。

This review turns post-vaccine mRNA therapeutics into a translation map: the hard bottlenecks are now delivery, immune silence, expression duration and repeat dosing.

这项研究让 Robin 把文献假说生成、湿实验数据分析和下一轮候选生成连成闭环，并在 dry AMD 中提出 ripasudil、KL001 和 ABCA1 线索。

This study connects literature-based hypothesis generation, wet-lab data analysis and follow-up candidate generation in Robin, yielding dAMD leads around ripasudil, KL001 and ABCA1.

这项研究把 Gemini 多智能体系统用于科学假说生成、辩论、排序和演化，并在 AML、肝纤维化和抗菌耐药机制中做了专家参与的实验验证。

This study uses a Gemini-based multi-agent system to generate, debate, rank and evolve scientific hypotheses, with expert-in-the-loop validation in AML, liver fibrosis and antimicrobial resistance.

这项研究把科学软件写作转化为可评分的 tree-search 问题，让 LLM 通过代码改写、外部研究想法和自动评分快速探索高质量解。

This study turns scientific software creation into a scorable tree-search problem, using LLM code rewrites, external research ideas and automatic scoring to explore high-quality solutions.

这项研究用瞬时 SMArT AND-gate reporter 在人 HSPC 中富集预期 HDR 编辑细胞，并同时降低潜在 genotoxic on-target 产物。

This study uses transient SMArT AND-gate reporters to enrich human HSPCs carrying the intended HDR edit while reducing potentially genotoxic on-target outcomes.

这篇 Review 把 AAV、LNP 和 EV 放在同一张工程与转化地图上，帮助判断不同基因治疗问题该匹配哪一种递送平台。

This review places AAVs, LNPs and EVs on one engineering and translation map to clarify which delivery platform fits which gene therapy problem.

这篇文章用 ProteinMPNN 重新设计进化过的 prime editor RT，在保留催化能力的同时提高稳定性、表达量和多种递送场景下的编辑效率。

这篇文章把 prime editing 的 pegRNA 3' 端稳定结构变成一个可筛选、可进化的工程模块，并证明它在细胞、原代细胞和小鼠体内都能提高编辑效率。