Model
AlphaGenome 2
Google DeepMind published AlphaGenome 2 on November 11, 2026. Trained on over 107 million cell profiles, it predicts DNA effects for specific cell types and is limited to research use.
Status Available Era Generative AI for everyone
- Access
- API
- Release date
- Developer
- Google DeepMind
What it is
AlphaGenome 2 is a research model from Google DeepMind for studying how cells read DNA. It predicts several cell-level readouts of DNA activity: gene expression, histone modifications, chromatin accessibility and, new in this version, DNA methylation.[1]
The model is conditional. It predicts for one tissue, cell type or state at a time, rather than giving a single answer for every kind of cell. The announcement's point is that the same DNA change can behave differently depending on where it occurs.[1]
Origins
Google DeepMind published AlphaGenome 2 on November 11, 2026. The announcement compares it with the original AlphaGenome, using a fine-tuned version of the original in each test.[1]
Training and scope
AlphaGenome 2 was trained on more than 107 million individual cell profiles from human and mouse data. The training data includes both bulk and single-cell measurements and covers 68 tissues.[1]
Key results
The announcement reports results on several tests. On the TenK10K dataset, which covers 28 primary immune cell types, DeepMind reports 30% for AlphaGenome 2 against 16% for the fine-tuned original model, measured on the identification of causal variants.[1]
For directional gene effects, which describe whether a DNA change raises or lowers a gene's activity, the company reports 43% for AlphaGenome 2 against 23% for the original model fine-tuned on single-cell data and for other specialized tools.[1] These figures are DeepMind's own reported results.
Access and limits
Non-commercial research can use the model today through the AlphaGenome API. Commercial research access is in private preview.[1]
The post says the model is intended for research use only. It has not been validated or approved for clinical or diagnostic purposes.[1]
Why it matters
The model gives researchers a way to ask how a change in DNA could affect gene activity in a specific tissue or cell type, instead of asking about cells in general. The reported results measure that ability on immune cells against the earlier model and other tools, while the access terms keep the work inside research settings.
20261
-
ReleaseGoogle DeepMind unveils AlphaGenome 2 for cell-type-specific DNA variant effects
On November 11, 2026, Google DeepMind unveiled AlphaGenome 2, along with an atlas of predicted effects for all 9 billion possible single-letter DNA changes in the human genome.
Developed by1
Notes and sources
- AlphaGenome 2: Deciphering how cells read DNA to better understand diseasehttps://deepmind.google/blog/alphagenome-2-deciphering-how-cells-read-dna-to-better-understand-disease/
Cite this page: AlphaGenome 2. TellWorld, https://www.tellworld.net/models/alphagenome-2. Updated 2026-11-15