Google DeepMind has announced AlphaGenome 2, a next-generation super intelligence (SI) model designed to predict how DNA sequences and their variations impact specific human cell types. This release marks a significant shift in SI applications for biology, moving from broad tissue averages to precise, cell-specific predictions to uncover cellular drivers of disease.
What Happened
AlphaGenome 2 is a conditional SI model that predicts DNA effects for one biological condition—such as a tissue, cell type, or state—at a time. Unlike its predecessor, which processed parallel outputs for bulk tissues, AlphaGenome 2 uses a target biological condition as input alongside a million-letter DNA sequence. This architectural change allows the model to learn how the same stretch of DNA is read differently across various cell types.
The model was trained on more than 107 million individual cell profiles across 68 tissues, including both bulk and single-cell data from human and mouse samples. It predicts core experimental readouts such as gene expression, histone modifications, and chromatin accessibility, while adding DNA methylation to its capabilities. AlphaGenome 2 is available for non-commercial research via the AlphaGenome API and in private preview for commercial research.
Why It Matters
This development represents a step-change in SI capability for human genetics. According to DeepMind, AlphaGenome 2 identifies nearly twice as many variants that change gene expression in cells compared to the previous state-of-the-art model. On the TenK10K dataset, the company reports that the SI system reliably identified 30% of causal variants and 43% of directional gene effects, compared to 16% and 23% respectively for the original AlphaGenome fine-tuned on single-cell data.
The model’s ability to generalize to new cell types without retraining is particularly impactful for researchers. Scientists can upload single-cell RNA-seq snapshots from just a few dozen cells, and AlphaGenome 2 can predict variant effects for unseen states, such as those occurring during disease or vaccination. For instance, the model distinguished vaccine-activated immune T cells from resting ones and revealed that a variant linked to severe COVID-19 disrupts gene activity more strongly in activated states.
By bridging human genetics and single-cell biology, AlphaGenome 2 helps researchers connect trait-associated variants from genome-wide association studies (GWAS) to specific cell types where disease unfolds. It extends enhancer-gene mapping to primary human cell types, such as brain astrocytes, and helps pinpoint causal variants and their target genes with higher confidence.
The Bottom Line
AlphaGenome 2 is intended for research use only and has not been validated for clinical or diagnostic purposes. DeepMind plans to release AlphaGenome 2 Skills in Google Antigravity, its agentic development platform, soon. The company is also collaborating with institutions like the Wellcome Sanger Institute to profile wider spectra of human cell types and states.
The source material does not specify exact pricing for the commercial private preview. Researchers are encouraged to submit use cases by November 30 for potential additional quota and credits.