DeepMind Unveils AlphaGenome: A New AI Tool for Unlocking Disease's Genetic Drivers
Google DeepMind has introduced AlphaGenome, an artificial intelligence tool designed to identify the genetic drivers of disease and facilitate the development of new treatments. This AI, which predicts how mutations influence gene regulation, has already been utilized by researchers—including at a hackathon focused on undiagnosed medical conditions—to analyze non-coding DNA variants.
AlphaGenome Overview
AlphaGenome's primary function is to predict how mutations alter gene regulation, specifically impacting when genes are activated, in which body cells, and at what expression levels. This capability targets the complex challenge of pinpointing specific genetic variations responsible for many common diseases, such as heart disease, autoimmune disorders, mental health conditions, and certain cancers, which have been linked to mutations affecting gene regulation.
The human genome consists of 3 billion base pairs of DNA.
Approximately 2% of this genome instructs cells to produce proteins, while the remaining 98% governs gene activity, providing instructions on where, when, and to what extent individual genes are expressed.
AlphaGenome is designed to understand the functional elements within this non-coding portion of the genome.
Researchers at DeepMind trained AlphaGenome using public databases of human and mouse genetics. This training enabled the AI to learn relationships between mutations in specific tissues and their impact on gene regulation. The tool can process up to 1 million DNA letters simultaneously and forecast how mutations will affect various biological processes.
Anticipated Applications and Development
The DeepMind team anticipates that AlphaGenome will assist scientists in mapping which genetic code strands are crucial for the development of particular tissues, such as nerve and liver cells. It is also expected to aid in identifying key mutations that drive cancer and other diseases. Furthermore, AlphaGenome could support new gene therapies by enabling researchers to design novel DNA sequences to activate specific genes selectively in certain cell types.
Natasha Latysheva, a DeepMind researcher, stated that AlphaGenome is intended to accelerate fundamental knowledge of the genetic code. The model was described in the journal Nature on January 28.
Application in Undiagnosed Disease Research
AlphaGenome was deployed during the Undiagnosed Hackathon, a three-day event held at the Mayo Clinic in Rochester, Minnesota. Over 100 researchers utilized the AI to address 29 undiagnosed medical conditions.
This event followed two previous hackathons held in Europe and was organized by the Wilhelm Foundation, a Swedish charity supporting families affected by undiagnosed rare diseases. The foundation was established by Helene and Mikk Cederroth, who experienced the loss of three of their four children to an undiagnosed condition. An estimated 350 million individuals worldwide live with an undiagnosed rare condition.
Current efforts to diagnose rare diseases typically focus on mutations within protein-coding regions of the genome (the exome).
Mutations in non-coding DNA sequences, which AlphaGenome is designed to interpret, pose particular difficulties for researchers and are frequently overlooked, according to Eric Klee, a bioinformatician at the Mayo Clinic and co-leader of the hackathon.
Before the September 2025 hackathon, Mr. Klee conducted a test to assess AlphaGenome's capacity to interpret non-coding variants. In this test, AlphaGenome's predictions regarding the effects of a variant previously linked to an individual's diagnosis supported experimental findings, which indicated that the mutation altered gene expression in cardiac cells but not in neural cells, aligning with the individual's symptoms.
Expert Perspectives
Carl de Boer, a researcher at the University of British Columbia not involved in the study, commented that AlphaGenome can identify if mutations affect genome regulation, which genes are impacted, how they are affected, and in what cell types. He suggested this capability could lead to the development of drugs to counteract these effects but noted that achieving models that negate the need for experimental confirmation will require ongoing scientific effort.
Marc Mansour, a clinical professor of pediatric hemato-oncology at UCL, described AlphaGenome as a "step change" in his work to find genetic drivers for cancer.
Statistical geneticist Gareth Hawkes from the University of Exeter emphasized the importance of AlphaGenome for understanding the 98% non-coding part of the genome, calling it a substantial advancement.
Helene Cederroth of the Wilhelm Foundation stated that individuals without a diagnosis are often disadvantaged.