Google DeepMind Unveils AlphaGenome 2, Trained on 100 Million Cell Profiles
Updated
Updated · The Verge · Nov 11
Google DeepMind Unveils AlphaGenome 2, Trained on 100 Million Cell Profiles
3 articles · Updated · The Verge · Nov 11
Summary
AlphaGenome 2 predicts how genetic variants affect individual cell types and cell states, giving researchers a more precise way to link DNA changes to disease mechanisms.
More than 100 million single-cell profiles across 68 tissues trained the model, which DeepMind says finds nearly twice as many gene-expression-altering variants as its previous state-of-the-art system.
The model also generalizes to new cell types and biological states without retraining and can flag gaps in experimental data, helping scientists decide what to test next in the lab.
Google is releasing AlphaGenome 2 for non-commercial research through the AlphaGenome API today, while commercial access starts in a limited private preview.
Its finer-grained predictions come with higher computing costs, so DeepMind says it will complement rather than replace AlphaGenome as AI labs push deeper into biology and drug discovery.
Will AlphaGenome 2's reliance on incomplete RNA data prevent it from ever truly simulating the complex reality of a living human cell?
Could DeepMind's new AI model rapidly accelerate targeted disease treatments, or will its massive computational demands hinder real-world medical breakthroughs?