Google DeepMind Open-Sources AlphaGenome to Advance DNA and Disease Research
DeepMind said the model can help researchers understand how variations in DNA affect these processes, as well as determine which genetic instructions are actively used by cells in different conditions.
Google DeepMind has open-sourced AlphaGenome, an artificial intelligence model designed to help researchers study DNA and better understand the biological processes behind health and disease.
The move expands access to a tool that was previously available only through a non-commercial API, which has already been adopted by more than 3,000 scientists and handles around one million requests a day.
AlphaGenome was built to accelerate DNA-focused medical research by analysing how genetic instructions influence protein production. Proteins play a central role in cellular function, and changes in how they are produced can contribute to disease.
DeepMind said the model can help researchers understand how variations in DNA affect these processes, as well as determine which genetic instructions are actively used by cells in different conditions.
One of AlphaGenome’s key advances is its ability to analyse DNA sequences of up to one million base pairs, far exceeding the context limits of earlier models. It also delivers higher-resolution predictions of molecular properties, offering more precise data for research.
The system combines three AI architectures: a convolutional neural network to analyse DNA base pairs, transformer models to refine the results, and a final neural layer that generates molecular property predictions.
In a paper published in Nature, DeepMind reported that AlphaGenome outperformed rival models in 25 of 26 internal benchmarks, while requiring relatively modest computing power, running on a single H100 GPU.
