DeepMind announces AlphaGenome 2 for cell-type DNA predictions
Original title: AlphaGenome 2: Deciphering how cells read DNA to better understand disease
The conditional model, trained on over 100 million cell profiles, is available for non-commercial use via an API.
Google DeepMind announced AlphaGenome 2, a conditional model that predicts how a DNA sequence and its variants affect one tissue, cell type, or cell state at a time. The lab says training covered more than 100 million single-cell profiles across 68 tissues and that outputs now include DNA methylation. Non-commercial research access is through the AlphaGenome API; commercial research is in private preview.
Unconfirmed: The recall and variant-count gains are DeepMind's own comparisons, not an independent evaluation.
- DeepMind says it finds nearly twice as many expression-changing variants as its prior model.
- On TenK10K immune cells, it reports 30% versus 16% causal-variant recall.
- A cell encoder can condition predictions on a scientist's own single-cell experiment.
- Inputs remain one-million-letter DNA sequences, now paired with a biological condition.
- Vendors
- Google DeepMind
- Products
- AlphaGenome API
- AI models
- AlphaGenome 2
From the original article
8 min read at deepmind.google
Predicting how DNA sequences and their variations impact specific human cell types and disease.
By Adam Gayoso, Natasha Latysheva, Pushmeet Kohli and Žiga Avsec · deepmind.google
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.