内容来源：https://blog.google/innovation-and-ai/technology/health/ai-health-research/

内容总结：

谷歌近日发布其在医疗健康领域的最新研究进展。过去五年间，谷歌团队发表了90篇经同行评审的疾病研究论文，累计被引用超过9万次，并与临床医生、学术机构和研究人员合作，推动科研成果从实验室走向患者。

在疾病机制解析方面，谷歌的AlphaGenome Atlas可预测人类基因组中单字母变异对分子生物学的影响，帮助研究人员从超过5.4万人的基因组数据中识别与衰老等相关的特定变异。AlphaFold被用于筛选3万种化合物，助力巴塞罗那全球健康研究所发现两种可用于治疗恰加斯病的已获批药物。谷歌还与哈佛大学合作，以纳米级分辨率重建了人脑碎片，并将数据开放共享。

在医疗工具开发方面，谷歌与斯坦福大学医学院合作，利用DeepVariant和DeepConsensus在不到8小时内完成全基因组测序诊断，创下吉尼斯世界纪录。公司还推出AlphaProtein Novo，可从零开始设计自然界不存在的新型酶，用于合成药物关键结构和降解有毒塑料。此外，谷歌地球AI模型结合环境、天气等数据，帮助公共卫生部门预测疾病暴发，已在刚果民主共和国用于更早识别埃博拉暴露和传播风险。

在临床支持方面，谷歌与伦敦帝国理工学院及英国国家医疗服务体系合作，对12.5万名女性进行乳腺筛查评估，该系统捕获了25%历史上被遗漏的间期癌症，并有望将专科医生阅片工作量减少40%。DeepSomatic在儿童白血病病例中检测出10个此前被遗漏的突变。谷歌流感预测模型在2025至2026流感季被美国疾控中心评为预测流感相关住院人数第一名，优于其他38个系统。

谷歌通过Google.org向全球相关机构提供超过6000万美元资助，支持利用AlphaFold等模型预测疟疾耐药突变、设计新型细胞疗法等研究。谷歌表示，通过在疾病各阶段介入，可将最棘手的健康挑战转化为可行解决方案，帮助人们更长寿、更健康地生活。

中文翻译：

应用技术来理解和治疗疾病
谷歌的健康研究与工具涵盖从基础细胞发现到实验室分析再到患者影响的各个层面。过去五年间，谷歌各团队发表了90篇关于理解和应对疾病的同行评审研究论文，累计获得超过90,000次引用。基于我们对技术如何加速科学进步以改善生活的愿景，我们正与临床医生、学术机构和研究人员合作，确保发现能够走出实验室、惠及患者。以下是我们如何助力推动医学研究和公共卫生解决方案。
解码疾病的运作机制
AlphaGenome Atlas是最全面的预测目录，可预测人类基因组中单字母变异如何影响分子生物学。
要有效治疗一种疾病，理解其在分子、遗传和细胞层面的运作方式大有裨益。我们的早期生物学模型作为科学家的协作研究工具，揭示自然界隐藏的机制，推动医学发现。

• 加速药物发现：巴塞罗那全球健康研究所的研究人员利用AlphaFold对30,000种化合物进行了虚拟筛选，以寻找针对恰加斯病的抗寄生虫活性。恰加斯病是一种被忽视的热带疾病，影响约800万人，若不治疗可能导致危及生命的并发症。这项研究帮助该研究所识别出两种FDA已批准、可重新用于恰加斯病的药物，加速了治疗时间。
• 绘制与衰老相关的遗传变异图谱：识别哪些罕见遗传变异与特定疾病相关颇具挑战，因为大量无影响的突变形成了“背景噪音”。埃克塞特大学的研究人员利用AlphaGenome Atlas搜索了超过54,000人的基因组数据，精确定位了驱动人体内关键循环蛋白丰度的特定变异，包括PLA2G7（与衰老相关）和EGLN1（一种重要的细胞氧传感器）。
• 在纳米尺度重建大脑：我们与哈佛大学利赫特曼实验室合作，以纳米级分辨率重建了一块人脑片段，并通过Neuroglancer公开了数据。我们还绘制了包括果蝇、斑马鱼幼体和鸣禽在内的模式生物的大脑图谱。这些神经图谱资源使研究人员能够推进对大脑功能的研究。
  加速医学解决方案的工具
  在AlphaFold支持下开发的蛋白质可视化图谱。
  实验室是生物发现被验证、工程化并转化为潜在医学和公共卫生解决方案的场所。我们的模型已取代缓慢的迭代流程，转而采用能够快速、准确处理大量数据的系统。
• 以超快速基因组测序支持遗传诊断：我们与斯坦福大学医学院合作，在临床试验中使用DeepVariant和DeepConsensus，帮助在时间紧迫的病例中确定遗传疾病的病因。该项目通过在全基因组测序中实现不到八小时的遗传诊断，协助实现了挽救生命的干预措施，这一前所未有的成就也获得了吉尼斯世界纪录的认可。
• 从零设计定制酶：我们推出了AlphaProtein Novo，这是一个机器学习流程，能够从第一性原理出发创造自然界不存在的新酶。在实验室测试中，它成功设计了定制酶来合成关键药物基序，还设计了其他酶在极端条件下降解有毒的环境塑料——展示了设计功能性蛋白质的能力，使其超越目前自然界已知的范围。
• 支持前瞻性公共卫生举措：一个社区的健康与其地理环境密切相关。我们将谷歌地球AI模型与环境、天气、出行及其他数据相结合，帮助公共卫生领导者预测疾病暴发并向脆弱社区提供医疗服务。在WHO非洲区域办事处和刚果（金）国家生物医学研究所的新研究中，对话式地理空间模型取代了数周的手动编码，帮助更早识别埃博拉暴露和传播风险——使当地应急人员能够主动部署移动实验室。
  以机器学习模型支持医疗提供者
  新的乳腺X光筛查系统有助于发现历史上被漏诊的间期癌症，并减少专科医生的阅片工作量。
  医学突破只有能帮助真实的人才有意义。我们的研究表明，机器学习模型可以帮助支持疾病检测，同时减轻行政工作负担，使医疗提供者能够专注于诊疗。
• 标记历史上被漏诊的癌症特征：在与伦敦帝国理工学院和英国国家医疗服务体系（NHS）合作发表于《自然·癌症》的两项研究中，我们在125,000名女性中评估了一套乳腺X光筛查系统。该系统发现了25%历史上被漏诊的间期癌症，同时显示可将专科医生阅片工作量减少多达40%——这可能为面临严重放射科医生短缺的医疗系统提供关键缓解。
• 揭示隐藏的儿童白血病突变：当Children's Mercy Hospital将其应用于此前已测序的真实世界儿童白血病病例时，DeepSomatic确认了所有已知变异，同时检测到10个此前遗漏的突变——尽管面临仅使用肿瘤样本的挑战。DeepSomatic已公开可用，有潜力为患者风险和靶向治疗的临床判断提供依据。
• 预测全球流感高峰：一个谷歌流感预测模型被美国疾病控制与预防中心（CDC）评为2025-2026季节预测流感相关住院人数的第一名。该模型使用我们的实证研究辅助工具开发，在准确性上超越了其他38个系统，与实际观察到的住院情况高度吻合，展示了该技术如何帮助公共卫生官员预判医疗服务需求并更好地为季节性疫情做好准备。
  让每个人都能获得更好的健康
  变革健康需要科学家、研究人员和机构携手合作。因此，我们为科学界提供先进的模型和开放获取平台，以推动突破性研究——同时提供他们应用这些研究所需的慈善资金和计算能力。
  通过Google.org，我们支持诸如马凯雷雷大学传染病研究所等组织，利用AlphaFold和AlphaMissense预测疟疾耐药突变；以及宾夕法尼亚大学，使用AlphaProteo和AlphaFold帮助设计新型细胞疗法，以增强对实体瘤和复杂疾病的治疗。2026年，Google.org已向全球各地正在应用先进模型变革我们理解和治疗疾病方式的组织提供了超过6,000万美元。
  通过在疾病的每个阶段加以应对，我们能够将最棘手的人类健康挑战转化为可实施的解决方案，帮助世界各地的人们活得更长、更健康。

英文来源：

Applying technology to understand and treat disease
Google’s health research and tools span from fundamental cellular discovery to laboratory analysis and patient impact. Over the last five years, Google teams have published 90 peer-reviewed research papers about understanding and addressing diseases, accumulating more than 90,000 citations. And building on our vision for how technology can accelerate science to improve lives, we’re collaborating with clinicians, academic institutes, and researchers to ensure discoveries make it out of the lab to benefit patients. Here is how our efforts are helping to advance medical research and public health solutions.
Decoding the mechanics of disease
AlphaGenome Atlas is the most comprehensive catalog of predictions for how single-letter variant changes in the human genome affect molecular biology.
To effectively treat an illness, it helps to understand how it operates at the molecular, genetic, and cellular levels. Our early-stage biology models act as collaborative research tools for scientists, revealing nature's hidden mechanics and advancing medical discoveries.

• Accelerating drug discovery: Researchers at the Institute for Global Health, Barcelona, used AlphaFold to virtually screen 30,000 compounds for antiparasitic activity against Chagas disease, a neglected tropical illness that affects roughly 8 million people and can lead to life-threatening complications if left untreated. This research helped the Institute identify two FDA-approved drugs that could be repurposed for Chagas, expediting time-to-treatment.
• Mapping genetic variants linked to aging: Identifying which rare genetic changes are linked to specific diseases is challenging because the volume of non-impactful mutations creates "background noise." Researchers at the University of Exeter used AlphaGenome Atlas to search through genomic data from over 54,000 individuals, pinpointing specific variants that drive the abundance of critical proteins circulating in the human body, including PLA2G7 (linked to aging) and EGLN1 (a vital cellular oxygen sensor).
• Reconstructing the brain at nanoscale: In partnership with Harvard’s Lichtman Lab, we reconstructed a human brain fragment at nanoscale resolution and made the data openly available via Neuroglancer. We additionally mapped the brains of model organisms including the fly, larval zebrafish, and songbird. These neural mapping resources enable researchers to advance their study of brain function.
  Accelerating tools for medical solutions
  Visual map of proteins developed with the support of AlphaFold.
  The lab is where biological discovery is validated, engineered, and turned into potential medical and public health solutions. Our models have replaced slow, iterative processes with systems that can process large amounts of data quickly and accurately.
• Supporting genetic diagnoses with ultrarapid genome sequencing: In collaboration with Stanford University School of Medicine, we used DeepVariant and DeepConsensus in a clinical trial to help determine the cause of genetic disease in time-critical cases. The program assisted in enabling life-saving interventions by achieving genetic diagnosis with whole genome sequencing in less than eight hours, an unprecedented achievement also recognized with a Guinness World Record.
• Designing custom enzymes from scratch: We introduced AlphaProtein Novo, a machine learning pipeline for creating new-to-nature enzymes from first principles. In lab tests, it successfully designed custom enzymes to synthesize key drug motifs, and others to degrade toxic environmental plastics under extreme conditions — demonstrating the ability to design functional proteins that can go beyond what is currently known in nature.
• Supporting proactive public health initiatives: A community's health is deeply connected to its geography. We are using our Google Earth AI models in combination with environmental, weather, mobility, and other data to help public health leaders forecast disease outbreaks and deliver care to vulnerable communities. In new research from WHO AFRO and the DRC’s INRB, conversational geospatial models replaced weeks of manual coding to help identify Ebola exposure and transmission risks sooner — allowing local responders to proactively deploy mobile laboratories.
  Supporting medical providers with machine learning models
  New mammography screening systems can help to catch historically missed interval cancers and reduce specialist reading workloads.
  A medical breakthrough only matters if it can help real people. Our research demonstrates that machine learning models can help support disease detection as well as alleviate the administrative workload so providers can focus on care.
• Flagging historically missed cancer signatures: In a pair of studies published in Nature Cancer in collaboration with Imperial College London and the UK’s National Health Service (NHS), we evaluated a mammography screening system across 125,000 women. The system caught 25% of historically missed interval cancers while showing the possibility of reducing specialist reading workloads by up to 40% — which could help provide crucial relief to healthcare systems facing severe radiologist shortages.
• Uncovering hidden pediatric leukemia mutations: When applied by Children’s Mercy Hospital to previously sequenced real-world childhood leukemia cases, DeepSomatic confirmed all known variants while detecting 10 previously missed mutations — despite the challenge of working with a tumor-only sample. DeepSomatic is publicly available and has the potential to inform clinical judgments around patient risk and targeted treatments.
• Predicting global flu surges: A Google flu forecasting model was ranked #1 by the Centers for Disease Control (CDC) for predicting flu-related hospital admissions during the 2025–2026 season. Developed using our Empirical Research Assistance tool, the model outperformed 38 other systems to accurately match observed hospitalizations, showing how the technology can help public health officials anticipate demand for medical services and better prepare for seasonal outbreaks.
  Better health within reach for everyone
  Transforming health requires scientists, researchers, and institutions working hand-in-hand. That’s why we equip the science community with the advanced models and open-access platforms to fuel breakthrough research — along with the philanthropic funding and computing power they need to apply that research.
  Through Google.org, we’re supporting organizations like the Infectious Disease Institute at Makerere University to forecast malaria drug resistance mutations with AlphaFold and AlphaMissense, and the University of Pennsylvania, using AlphaProteo and AlphaFold to help design novel cellular therapies that can augment the treatment of solid tumors and complex diseases. In 2026, Google.org has provided over $60 million to organizations around the world that are applying advanced models to transform how we understand and treat diseases.
  By meeting disease at every stage, we can turn the most intractable human health challenges into actionable solutions, helping people everywhere live longer, healthier lives.