Google DeepMind 发布 AlphaGenome Atlas 基因组预测图谱

谷歌DeepMind推出了一款人工智能工具，其科学家声称这款工具能够帮助揭开人类基因组的奥秘，改变我们对生物学的理解，加速科学研究，并最终为疾病的新疗法铺平道路。

研究人员在周二发布的一篇博客文章中表示，这一平台名为AlphaGenome Atlas，包含“人类基因组中每个可能的DNA字母变异的预测图谱”：https://deepmind.google/blog/alphagenome-atlas-a-predictive-map-of-every-possible-dna-letter-change-in-the-human-genome。

DNA是由四个化学“字母”组成的字母表书写的——通常简称为A、C、G和T——人类基因组大约包含三十亿个字母对。这些字母包含了使生命运作的指令，例如基因何时、在何处以及以何种程度被开启或关闭。单个字母的变化可能无害，也可能造成个体之间的普通差异，或者在疾病中起作用——一个主要挑战是弄清楚哪些变化重要以及如何影响。由于大约有九十亿种可能的单字母替换，这是一个艰巨的任务。

Atlas包含对这九十亿种变异如何在分子水平影响身体的预测，例如改变某种蛋白质的产生量。研究人员称其为“最全面的基因突变对分子生物学影响的目录”。谷歌表示，科学家可以通过网页门户、其代理开发平台Antigravity中的技能以及通过AlphaGenome界面探索这些预测。

为了帮助研究人员筛选数十亿种可能性，并专注于值得更深入研究的突变，谷歌还发布了称为变异影响评分（Variant Impact Score，简称VIA）的工具，该工具借助公司其他模型来预测DNA变化的效果。“现在，研究人员可以快速对变异进行排名，同时解释它们的分子效应，”公司博客写道。

该项目建立在 AlphaGenome 的基础上：https://deepmind.google/blog/alphagenome-ai-for-better-understanding-the-genome/，这是 DeepMind 去年推出的一个 AI 模型，旨在帮助科学家识别疾病的遗传驱动因素，以及 AlphaMissense：https://deepmind.google/blog/a-catalogue-of-genetic-mutations-to-help-pinpoint-the-cause-of-diseases/，这是一个早期工具，专注于预测哪些小突变可能改变蛋白质。Atlas 的能力更为强大，扩展了对整个基因组的预测，包括绝大多数不直接编码蛋白质的片段，而这些片段则可能控制基因的表达方式。

在一次新闻发布会上，DeepMind 基因组学负责人 Ziga Avsec 承认基础模型——AlphaGenome——已经发布，但表示将其能力转化为全基因组目录需要时间。他说：“基本上我们花了一些时间来真正预计算并分析这么多变体，因为这个空间太大了。”

AlphaGenome 使用人类和小鼠基因组的公共数据库进行训练，从而学会 DNA 变化与生物过程之间的模式。将这些预测应用于数十亿种可能的变体，产生了一个 Google 称大约为 1 petabyte（拍字节）大小的庞大数据集。

公司表示，从今天起，它将通过网站向研究人员提供 Atlas 供非商业使用，并将在“近期”通过 Google Cloud 提供商业使用。

Atlas 是谷歌一系列利用人工智能解决科学和医学核心问题的最新尝试之一，此时 DeepMind 联合创始人 Demis Hassabis 从管理 AI 实验室的工作中退居二线，专注于科学研究：/podcast/979370/google-deepmind-ai-race-lose-jeff-dean-demis-hassabis，包括领导药物研发衍生公司 Isomorphic Labs。该公司在这一领域最知名的工作是 AlphaFold，一种蛋白质结构预测模型，使 Demis Hassabis 和 John Jumper 赢得了 2024 年诺贝尔化学奖：https://www.nobelprize.org/prizes/chemistry/2024/popular-information/。该公司还开发了用于天气预测的 AI 工具：/tech/988921/weather-forecast-ai-model-google-satellite-update，优化计算和数学中寻找新解决方案的能力：/news/666377/googles-says-its-new-ai-agent-can-find-new-solutions-in-computing-and-math，并通过代理式“合作科学家”帮助研究人员：https://deepmind.google/blog/co-scientist-a-multi-agent-ai-partner-to-accelerate-research/。

Google DeepMind has unveiled an AI tool that its scientists claim could help unravel the mysteries of the human genome and transform our understanding of biology, accelerating scientific research and ultimately paving the way for new treatments for diseases.

The platform, called AlphaGenome Atlas, contains a “predictive map of every possible DNA letter change in the human genome,” the researchers said in a blog post：https://deepmind.google/blog/alphagenome-atlas-a-predictive-map-of-every-possible-dna-letter-change-in-the-human-genome published on Tuesday.

DNA is written in an alphabet of four chemical “letters” — usually shortened to A, C, G, and T — and the human genome contains roughly three billion letter pairs. Those letters contain the instructions that make life work, such as how, when, and where genes are switched on and off, and to what degree. Changes to individual letters can be harmless, contribute to ordinary differences between people, or play a role in disease – a major challenge is figuring out which changes matter and how. It is a tough task as there are roughly nine billion potential single-letter substitutions.

Atlas contains predictions for how each of these nine billion variants could affect the body at a molecular level, such as changing how much of a particular protein is produced. The researchers call it “the most comprehensive catalogue of how genetic mutations affect molecular biology.” Google says scientists can explore these predictions through a web portal, as a skill in its agentic development platform Antigravity, and through its AlphaGenome interface.

To help researchers sift through the billions of possibilities and focus on the mutations that warrant closer attention, Google is also releasing what it calls a Variant Impact Score (AVI), that draws on the company’s other models for predicting the effects of DNA changes. “Now, researchers can rapidly rank variants and interpret their molecular effects at the same time,” the company’s blog said.

The project builds on AlphaGenome：https://deepmind.google/blog/alphagenome-ai-for-better-understanding-the-genome/, an AI model DeepMind unveiled last year to help scientists identify the genetic drivers of disease, as well as AlphaMissense：https://deepmind.google/blog/a-catalogue-of-genetic-mutations-to-help-pinpoint-the-cause-of-diseases/, an earlier tool focused on predicting which small mutations might alter proteins. Atlas goes much further, extending predictions across the genome, including the vast majority of stretches that do not directly code for proteins, but can instead control how genes behave.

In a press briefing, Ziga Avsec, DeepMind’s genomics lead, acknowledged that the underlying model — AlphaGenome — had already been released, but said turning its capabilities into a genome-wide catalog took time. “Basically it took us some time to really precompute and also analyze this many variants because the space is so big,” he said.

AlphaGenome was trained using public databases of human and mouse genomes, allowing it to learn patterns between DNA changes and biological processes. Applying those predictions to billions of possible variants produced a massive dataset that Google says is roughly 1 petabyte in size.

The company says it is making Atlas available to researchers for noncommercial use through its website starting today, and for commercial use on Google Cloud “soon.”

Atlas is the latest in a string of efforts from Google to use AI to tackle core problems in science and medicine, coming at a time when DeepMind cofounder Demis Hassabis steps back from running the AI lab to focus on scientific research：/podcast/979370/google-deepmind-ai-race-lose-jeff-dean-demis-hassabis, including leading drug-discovery spinoff Isomorphic Labs. The company’s best-known work in this area is AlphaFold, the protein-structure prediction model that won Demis Hassabis and John Jumper the 2024 Nobel Prize in Chemistry：https://www.nobelprize.org/prizes/chemistry/2024/popular-information/. The company has also developed AI tools for predicting the weather：/tech/988921/weather-forecast-ai-model-google-satellite-update, optimizing finding new solutions in computing and mathematics,：/news/666377/googles-says-its-new-ai-agent-can-find-new-solutions-in-computing-and-math and assisting researchers through an agentic “co-scientist：https://deepmind.google/blog/co-scientist-a-multi-agent-ai-partner-to-accelerate-research/.”