Google DeepMind has introduced AlphaGenome, a new artificial intelligence model that could fundamentally transform our understanding of DNA—the complete set of instructions for building and operating the human body. According to researchers, this tool has the potential to transform disease research and the discovery of new drugs.
The model was described in the prestigious journal Nature, and since being made available for non-commercial use, it has already been used by 3,000 scientists around the world. Natasha Latysheva, a research engineer at DeepMind, describes AlphaGenome as a tool for understanding functional elements in the genome that should accelerate fundamental understanding of the code of life.
What AlphaGenome Can Do
The human genome consists of three billion letters of DNA code, represented by the letters A, C, G, and T. Only 2% consists of genes that encode all the proteins needed for the body to grow and function. The remaining 98% is referred to as the "dark genome"—a less-explored area that plays a key role in organizing how genes are used and where many disease-associated mutations are found.
AlphaGenome can analyze up to one million letters of code at a time, helping to accelerate the discovery of the dark genome's secrets. The model predicts where genes are located, as well as how the dark genome affects their expression (whether a gene is highly active or suppressed) and gene splicing—the process by which the body creates different proteins from a single gene.
Crucially, the model can predict the impact of changing a single letter in the genetic code. This capability opens the door to understanding why small differences in our DNA increase the risk of conditions such as high blood pressure, dementia, or obesity.
Technical Capabilities
AlphaGenome takes a long DNA sequence—up to one million base pairs—as input and predicts thousands of molecular properties characterizing its regulatory activity. The training data came from large public consortia such as 4D Nucleome and FANTOM5, which experimentally measured these properties covering important modes of gene regulation across hundreds of human and mouse cell and tissue types.
The model achieves state-of-the-art performance across a wide range of genomic benchmarks. When generating predictions for individual DNA sequences, AlphaGenome outperformed the best external models in 22 out of 24 evaluations. When predicting the regulatory effect of a variant, it matched or surpassed the best external models in 24 out of 26 evaluations.
Research Applications
Dr. Gareth Hawkes of the University of Exeter is using AlphaGenome to explore how mutations may alter our risk of obesity and diabetes. Studies that sequenced the entire genetic code of tens of thousands of people identified variants associated with these conditions, but they are often located in the dark genome. "These variants directly affect some important piece of biology that we really don't understand," Hawkes told the BBC. Using AlphaGenome allows researchers to quickly predict what these variants do so they can be tested in the laboratory. "I wouldn't say that the dark side of the genome has been solved by AlphaGenome, but it is a major leap. I'm really excited."
Another area in which the model could accelerate research is cancer. AlphaGenome has been used to predict which mutations drive cancer and are also potential treatment targets, and which mutations occur by chance. For example, the team used AlphaGenome to investigate a potential mechanism of a cancer-associated mutation in patients with T-cell acute lymphoblastic leukemia (T-ALL).
Expert Assessments
Dr. Robert Goldstone, head of genomics at the Francis Crick Institute, described the model as a "significant milestone in genomic AI" and the breakthrough as an "incredible technical achievement" because of its "ability to predict gene expression from DNA sequence alone."
Prof. Ben Lehner, head of generative and synthetic genomics at the Wellcome Sanger Institute, said they had tested AlphaGenome in more than half a million experiments and that it performs very well. However, he cautioned that it is "far from perfect" and that much work remains to be done.
Dr. Caleb Lareau of Memorial Sloan Kettering Cancer Center called AlphaGenome a milestone for the field: "For the first time, we have a single model that unifies long-range context, base-level precision, and state-of-the-art performance across the full spectrum of genomic tasks."
Deployment Options and Limitations
Latysheva expressed excitement about the model's potential to identify which mutations cause disease and help determine the causes of rare genetic disorders. The model could be used to "add another piece to the puzzle of discovering drug targets and ultimately developing new drugs." In the long term, it could also be used in synthetic biology and the design of new DNA sequences that could be used in gene therapies.
However, the DeepMind team acknowledges the current limitations. The model is less accurate in some areas, such as predicting how genes are regulated over long distances (more than 100,000 letters of code). The team also wants to improve the model's accuracy across different tissues—for example, a neuron in the brain has the same genetic code as a beating heart cell, but each has different properties based on how the genetic instructions are used in each cell type.
Pushmeet Kohli, vice president of science and strategic initiatives at Google DeepMind, believes we are at the beginning of a new era of scientific progress: "I think AI will enable a range of different breakthroughs." The DeepMind team received the 2024 Nobel Prize in Chemistry for its work on AlphaFold—an AI system that predicts the 3D structure of proteins in the body.
AlphaGenome is now available for non-commercial use through an API, and researchers around the world are invited to share potential use cases and feedback through the community forum.
