Google DeepMind has released AlphaGenome 2, a conditional DNA sequence model designed for noncommercial research purposes. The model predicts the effects of genetic variants on specific human cell types. DeepMind plans to integrate the system into the Google Antigravity agentic platform.
The model was trained on over 107 million genetic profiles and can infer missing data without retraining. On the TenK10K benchmark, AlphaGenome 2 recovered 30% of causal variants, nearly double the 16% achieved by the previous version. DeepMind precomputed predictions to support drug development and GWAS analysis.
Researchers can use the tool to identify disease-relevant gene changes that were previously difficult to detect. The system adapts to new experiments and helps scientists understand complex genetic interactions. This advance makes it easier to pinpoint the genetic roots of rare diseases.
