Google DeepMind unveils AlphaGenome AI to decode DNA effects

Google's DeepMind division has launched AlphaGenome, a new AI model designed to predict how small changes in DNA affect gene activity and molecular processes. This breakthrough builds on the success of AlphaFold, DeepMind's earlier AI tool for protein folding, which earned a Nobel Prize in 2024.

Understanding the Genome

• The human genome contains 3 billion genetic letters, but their functions remain largely unknown.
• AlphaGenome aims to answer fundamental questions about how DNA variations influence gene expression and health.
• The model can predict whether a genetic mutation will increase or decrease gene activity, a task traditionally requiring lab experiments.

Potential Applications

• Cancer Research: AlphaGenome could identify rare mutations driving cancer, helping doctors target treatments.
• Rare Diseases: The AI may diagnose undetermined genetic conditions by analyzing DNA alterations.
• Virtual Experiments: Researchers like Caleb Lareau of Memorial Sloan Kettering Cancer Center say the tool could replace some lab work with computer simulations.

Availability and Future Plans

• AlphaGenome will be free for noncommercial use, with plans to offer it to biotech companies later.
• Google is exploring ways to expand its commercial applications, similar to its spinout Isomorphic Labs, which focuses on AI-driven drug discovery.

Limitations

• The model does not predict personal traits or ancestry like 23andMe.
• It focuses on molecular-level insights rather than individual health predictions.

Broader Vision

DeepMind CEO Demis Hassabis envisions a future where AI can simulate an entire virtual cell, with AlphaGenome serving as a stepping stone. Pushmeet Kohli, DeepMind's VP of Research, calls it a "milestone" toward understanding the "semantics of DNA."

For more on AlphaFold, read this article.