AlphaGenome: The AI That Could Transform Genetic Testing and Drug Discovery

Google DeepMind's AlphaGenome is a new AI system designed to predict how DNA variants affect gene regulation and other biological processes. Its developers say the model could sharpen genetic diagnosis, accelerate drug-target discovery and even help design engineered DNA sequences for therapies.

What AlphaGenome Does

AlphaGenome was trained on both human and mouse genomic data and predicts the functional consequences of DNA mutations across thousands of biological signals. The team reports the model can simultaneously forecast 5,930 human or 1,128 mouse genetic signals, providing detailed, variant-level insights into gene regulation and cellular function.

Performance

Published in Nature, the model matched or outperformed current state-of-the-art tools in 25 of 26 benchmark evaluations. According to the authors, these results show AlphaGenome often offers improved accuracy for interpreting non-coding regions of the genome—areas that have been historically hard to decode.

Applications and Potential

Researchers and DeepMind engineers highlight several important uses:

• Drug Discovery: Combining AlphaGenome predictions with large genetic-association datasets (for example, UK Biobank) can help pinpoint genes and cell types linked to particular traits, aiding target selection.
• Clinical Genetics: The model can help distinguish likely causal driver mutations from passenger mutations, improving variant prioritisation in cancer and rare-disease diagnosis.
• Gene Therapy and Design: A robust sequence-to-function model could enable design of DNA sequences with cell-type specific activity—for example, activating a gene in neurons but not muscle cells.

“By combining large genetic association studies with AlphaGenome predictions, scientists could better pinpoint the genes and cell types associated with a particular trait or disease,” said Natasha Latysheva, research engineer at Google DeepMind.

Adoption and Availability

AlphaGenome has been available to researchers via an API since June 2025. Pushmeet Kohli, vice president of science and strategic initiatives at Google DeepMind, reported roughly 3,000 scientists have made around one million API calls from 160 countries. DeepMind is releasing the model and pricing for non-commercial research use while testing a commercial edition.

Major academic labs—including UCL and the Wellcome Sanger Institute—are using or validating the tool. The Wellcome Sanger Institute validated AlphaGenome against about 500,000 new experiments and reported strong performance in internal tests.

Context and Cautions

AlphaGenome builds on recent advances in biological AI—DeepMind founders and leaders were recognised in 2024 for their protein-folding work—but experts emphasise limits. Model performance depends heavily on the quantity, quality and standardisation of training data. As Professor Ben Lehner of the Wellcome Sanger Institute noted, current biological datasets are often small or inconsistent, and generating larger, standardised datasets is a top priority for the next generation of genomic AI.

“AlphaGenome is not a magic bullet for all biological questions, but it is a foundational, high-quality tool that turns the static code of the genome into a decipherable language for discovery,” said Dr Robert Goldstone of the Francis Crick Institute.

Looking Ahead

AlphaGenome represents a major step toward reading the regulatory "grammar" of DNA and translating the human genome into actionable biological insight. Its near-term impact is likely to be strongest in research settings—improving variant interpretation, accelerating target discovery and enabling experimental design—while broader clinical and commercial applications will depend on further validation, larger datasets and careful integration with experimental biology.