DeepMind’s AlphaGenome Predicts DNA Mutation Impact with AI | Nature Publication
- Researchers at Google DeepMind have unveiled January 28, 2026, a new artificial intelligence tool called AlphaGenome, designed to predict the impact of mutations in human DNA.
- For decades, scientists have understood that the vast majority of our DNA – over 98% – doesn’t directly code for proteins.
- AlphaGenome distinguishes itself from previous models, such as Enformer (announced in 2021), by its ability to analyze significantly longer stretches of DNA – up to 1 million base...
Researchers at Google DeepMind have unveiled , a new artificial intelligence tool called AlphaGenome, designed to predict the impact of mutations in human DNA. Building on the success of their earlier AI, AlphaFold – which revolutionized protein structure prediction and earned its developers a share of the Nobel Prize in Chemistry – AlphaGenome aims to decipher the complexities of the non-coding regions of the human genome.
For decades, scientists have understood that the vast majority of our DNA – over 98% – doesn’t directly code for proteins. This so-called “junk DNA” was initially dismissed, but is now recognized as crucial for regulating gene activity and influencing health and disease. Understanding how variations within this non-coding DNA affect gene expression has remained a significant challenge. AlphaGenome is designed to address this very problem.
AlphaGenome distinguishes itself from previous models, such as Enformer (announced in ), by its ability to analyze significantly longer stretches of DNA – up to 1 million base pairs at a time. This expanded capacity allows the AI to identify relationships between genetic changes and gene activity even when those changes occur at a distance from the gene itself. For example, the tool can predict whether a mutation in a DNA sequence acting as an enhancer (a “gene switch”) will activate or deactivate a gene.
The model was trained using publicly available human and mouse genome data, learning to predict how DNA sequences influence a range of biological processes. Specifically, AlphaGenome can simultaneously predict the effects of 11 types of mutations related to functions like gene expression and splicing – the process of selecting specific parts of a DNA sequence. Researchers report that AlphaGenome can distinguish and predict the characteristics of 5,930 human genetic signals and 1,128 mouse genetic signals.
In testing, AlphaGenome demonstrated superior performance compared to existing prediction models. Researchers found that in 25 out of 26 comparisons, AlphaGenome’s predictions were equal to or better than those of current state-of-the-art models. One example highlighted by researchers involved the TAL1 gene, which plays a role in immune cell function. AlphaGenome accurately predicted the effect of a mutation located approximately 8,000 base pairs away from the gene, a mutation that can lead to uncontrolled immune cell proliferation and leukemia.
While the development of AlphaGenome is being hailed as a significant advancement, experts caution that clinical applications are still some time away. Dr. Peter Gu, a computational biologist at Cold Spring Harbor Laboratory, described the achievement as a “phenomenal engineering achievement” but emphasized the ongoing need for refinement.
Researchers acknowledge that the model’s predictive power diminishes as the distance between a mutation and the affected gene increases. Some scientists have raised concerns about the reliance on data used for training the model. Dr. Stephen Salzberg of Johns Hopkins University cautioned against over-optimism, while Dr. Catherine Pollard of the Gladstone Institute pointed out that AlphaGenome currently predicts the effect of a single mutation in a single genome, and isn’t yet equipped to analyze a patient’s entire genome for potential health threats.
Despite these limitations, AlphaGenome is already being utilized by researchers in various fields. Professor Mark Mansour, a leukemia researcher at University College London, described the tool as “groundbreaking” in the search for the genetic causes of cancer. However, he also stressed the importance of interpreting the results cautiously.
DeepMind has made AlphaGenome available in preview form through an API, allowing researchers worldwide to access the tool for non-commercial research purposes. As of , over 3,000 scientists from 160 countries have begun using AlphaGenome to investigate diseases including cancer, infectious diseases, and neurodegenerative disorders.
The researchers plan to expand the scope of AlphaGenome by incorporating data from additional biological species and further refining its ability to predict the function of non-coding DNA regions. The research detailing AlphaGenome’s development was published as a cover article in the journal Nature on .