Google DeepMind announced AlphaGenome 2 in a blog post, describing it as a next-generation DNA sequence model.confirmed The system predicts how genetic variations impact specific human cell types, moving beyond broad tissue averages.The new model helps scientists see how small DNA changes affect specific cell types, not just whole tissues.The conditional architecture enables fine-grained prediction of variant effects on single-cell gene activity and chromatin accessibility.
Evidence: confirmed
AlphaGenome 2: Deciphering how cells read DNA to better understand diseaseaifleet-deepmind
What was released or claimed
DeepMind stated the model bridges single-cell biology and human genetics. It is available for non-commercial research via an API and in private preview for commercial use. The company claims it turns static cell atlases into a predictive engine.company claim
Evidence: company claim
AlphaGenome 2: Deciphering how cells read DNA to better understand diseaseaifleet-deepmind
Who it affects
The tool targets scientists studying disease drivers. It helps researchers prioritize variant effects in rare diseases and complex traits. The model serves those mapping gene expression for millions of cells.
What is still unknown
The material does not specify validation metrics or error rates. It does not detail computational requirements for local deployment. No independent verification of the performance claims is provided in the text.
So what
By resolving activity inside specialized cells, the model may uncover DNA variants acting only in specific contexts. This could help identify cellular drivers of disease that broad-tissue models miss.
What the paper shows
AlphaGenome 2 tests variant effects on single-cell gene activity and chromatin accessibility.reported It reports nearly double the variant identification of the previous model.reported The authors note the conditional design allows cross-modality imputation for missing data.
Evidence: reported
AlphaGenome 2: Deciphering how cells read DNA to better understand diseaseaifleet-deepmind
Why it matters
- The conditional architecture allows imputation of unmeasured properties like chromatin accessibility from single measurements.
- Training on over 100 million cell profiles enables resolution of activity in distinct heart muscle or blood immune cells.
- The model predicts effects for cell and disease states not seen during training, expanding generalization capabilities.
- Researchers can prioritize experiments by predicting which variants change gene expression in specific cell types.
Topics: models, genomics, Google DeepMind, single-cell biology, AI research
Updated
Then / now / next
The original AlphaGenome predicted variant effects across broad tissues using parallel prediction.
AlphaGenome 2 uses a conditional architecture to predict effects for specific cell types and states.
The model is available for non-commercial research via an API and in private preview for commercial research.
Story timeline
- Then The original AlphaGenome predicted variant effects across broad tissues using parallel prediction.
- 11 Nov 2026 AlphaGenome 2: Deciphering how cells read DNA to better understand disease
- Now AlphaGenome 2 uses a conditional architecture to predict effects for specific cell types and states.
- Next The model is available for non-commercial research via an API and in private preview for commercial research.
The strongest doubt
Counterpoint from the sources
The material presents only DeepMind's performance claims without independent validation or comparison to other recent models.
Ask the article
Local keyword search of this article; no generated answers.Which of these is confirmed?
Check the answer
Google DeepMind announced AlphaGenome 2 in a blog post, describing it as a next-generation DNA sequence model.
The matched primary source and announcement wording support the confirmed label.
Did this article give you the evidence you needed?
Your choice stays on this device. No public tally.Questions readers ask
How does AlphaGenome 2 differ from the original AlphaGenome?
It uses a conditional model to predict effects for one biological condition at a time, rather than parallel prediction across tissues.
What data was used to train the model?
The model was trained on bulk and single-cell data from human and mouse samples, including over 107 million individual cell profiles.
Is the model available for commercial use?
It is available for non-commercial research via an API and in private preview for commercial research.
Connected stories
Reading receipt
- The model identifies nearly twice as many gene expression variants as its predecessor.
- It uses a conditional architecture to process one biological condition at a time.
- Training data includes over 107 million individual cell profiles from human and mouse samples.
Still open: How does the model perform on cell types with limited training data?
Still open: What are the computational costs for large-scale variant analysis?
