人类基因组在大约二十年前首次被绘制完成，但其中仅有2%包含制造蛋白质的指令，而其余98%长期以来被称为「垃圾」DNA。如今生物学研究显示，这些非编码DNA中的变异与癌症、糖尿病及心脏病等重大疾病密切相关，Imperial College London的Alessia David指出其绝非无用之物。

为应对这一挑战，Google旗下DeepMind开发了AI模型AlphaGenome，并预先计算了人类基因组中全部90亿个可能的单字母突变，构建出容量达1 PB（1-petabyte）的AlphaGenome Atlas资料库。该资料库使研究人员能在数秒内批次检索数千至数百万个变异的预测结果，显著加速了对疾病成因的遗传学研究。

DeepMind副总裁兼Google Cloud首席科学家Pushmeet Kohli表示，该工具将对学术界免费开放，但商业用户（包括同系药物研发公司Isomorphic Labs）则需支付授权费。尽管联合创始人Demis Hassabis于八月卸任执行长引发了外界对基础研究投入的讨论，但团队深信此类突破将持续为商业应用与人类遗传学带来深远影响。

Scientists mapped the human genome roughly two decades ago, yet only 2% encodes instructions for protein synthesis, leaving the remaining 98% historically labeled as junk DNA. Modern biological research has now linked variations within these non-coding regions to complex conditions including cancer, diabetes, and heart disease, confirming, as Imperial College London's Alessia David observed, that this sequence is not junk at all.

Addressing this challenge, Google DeepMind introduced the AI model AlphaGenome and released the AlphaGenome Atlas, a 1-petabyte (1-PB) database containing precomputed predictions for all 9 billion possible single-letter alterations in the human genome. This repository enables scientists to query thousands or millions of variants simultaneously in seconds, substantially accelerating research into disease origins and genetic predispositions.

Pushmeet Kohli, vice president of research at DeepMind and chief scientist at Google Cloud, noted that while the atlas remains freely accessible to academic researchers, commercial entities such as sister company Isomorphic Labs will pay licensing fees. Although cofounder Demis Hassabis stepped down as chief executive officer in August, DeepMind maintains that these fundamental breakthroughs will yield significant commercial and genetic insights.