Google’s AlphaGenome 2 zooms in on the cellular roots of disease

Sundar Pichai confirmed the non-commercial access, according to Traders Union.

The model is designed to predict how DNA sequences and their variations affect particular human cell types and states, which Google DeepMind says can help scientists find the cellular drivers of disease. It builds on last year's AlphaGenome, which predicted variant effects across broad tissues, and on AlphaGenome Atlas, a platform holding predictions for all 9 billion possible single-letter changes in the human genome.

AlphaGenome 2 shifts from broad tissue averages to individual cell types and states. Google DeepMind says it identifies nearly twice as many variants that change gene expression in cells compared with its previous state-of-the-art model. The model was trained on more than 100 million individual cell profiles across 68 tissues, drawing on both bulk and single-cell data from human and mouse samples.

Rather than predicting in parallel, the model works conditionally, taking a target biological condition, defined as a tissue, cell type or state, as input alongside a million-letter DNA sequence. It predicts gene expression, histone modifications and chromatin accessibility, and adds DNA methylation. It can also infer missing experimental data: given one measurement such as single-cell gene activity, it can predict unmeasured properties like chromatin accessibility.

The model can generalize to new single-cell experiments without retraining. Scientists can upload a single-cell RNA-seq snapshot from just a few dozen cells, and Google DeepMind says the model can match or outperform models trained or fine-tuned directly on those datasets. In an example drawn from a COVID-19 mRNA vaccine study, AlphaGenome 2 distinguished vaccine-activated immune T cells from resting ones and found that a variant linked to severe COVID-19 disrupts gene activity more strongly in the activated state, an effect that would be missed in unactivated cells.

Google DeepMind also precomputed predictions for trait-associated variants across hundreds of human cell and tissue conditions, aimed at connecting variants from genome-wide association studies with single-cell maps of the body. The company said it is collaborating with researchers at the Wellcome Sanger Institute to profile more human cell types, developmental lineages and perturbed cellular states.

The caveat is scope of use. Google DeepMind states that AlphaGenome and AlphaGenome 2 are intended for research use only and have not been validated or approved for clinical or diagnostic purposes. The company also said AlphaGenome 2 Skills will come soon in Google Antigravity, its agentic development platform, alongside Gemini 4.