Google DeepMind Releases AlphaGenome 2 for Cell-Type-Specific Variant Effects
By Asif Razzaq β November 11, 2026
Google DeepMind has released AlphaGenome 2, a DNA model that predicts variant effects within specific cell types. AlphaGenome 2 is a conditional sequence-to-function model: it reads 1 Mb of DNA alongside a target cell state. Its predecessor averaged signal across bulk tissues; the new model resolves closely related cell types and adapts to unseen single-cell datasets without fine-tuning. It is available through the AlphaGenome API for non-commercial research.
TL;DR
- Size: Input is a 1 Mb (1,048,576 bp) DNA window, for human or mouse genomes.
- Runs on: Hosted API only for now. Training used 2,048 TPU v4 chips; model weights will be released "upon final publication."
- Performance: Outperforms fine-tuned AlphaGenome 1, Borzoi, and Scooby on single-cell eQTL and enhancerβgene benchmarks.
- Best result: 43% recall at 90% sign accuracy on TenK10K sc-eQTLs, versus 23% for fine-tuned AlphaGenome 1.
- Worst result: Outside accessible chromatin, sign-prediction AUROC reaches only 0.63.
- Bottom line: Strongest at cell-state conditioning from raw scRNA-seq. Weakest on practical fronts: no local weights and non-commercial use only.
What Is AlphaGenome 2?
AlphaGenome 2 is a U-Net st...
via MarkTechPost