By By Akhtar Saleem Khan
Head of IT | The BNN Media – The Best News Network
News Beyond the Borders
Google DeepMind has unveiled AlphaGenome 2, a new artificial intelligence model that reads DNA the way individual cells do — predicting how a single genetic change plays out in one specific type of human cell, and offering scientists a finer tool for tracing the cellular roots of disease.
London, United Kingdom — Google DeepMind announced AlphaGenome 2 on Tuesday, 11 November 2026, its next-generation DNA-sequence model that predicts how DNA sequences and their variations affect specific human cell types, to uncover the cellular drivers of disease.
Unlike the first AlphaGenome, which predicted variant effects across broad tissues, AlphaGenome 2 is a conditional model: it takes a specific biological condition — a tissue, cell type or cell state — alongside a 1-million-letter DNA sequence.
From Tissues to Single Cells
Every cell in the body carries the same DNA blueprint, yet each cell type switches different genes on or off to do its job — a heart muscle cell reads the same genome differently from an immune cell fighting an infection. Global efforts like the Human Cell Atlas are mapping gene expression for millions of cells, but predicting how subtle DNA variations affect that expression in individual cells has remained an unsolved challenge.
The first AlphaGenome, launched last year, predicted how DNA variants affect molecular activity across broad tissues, and DeepMind later expanded its reach with the AlphaGenome Atlas, a platform containing predictions for all 9 billion possible single-letter changes in the human genome. AlphaGenome 2 zooms in further — from broad tissue averages to individual cell types and cell states — and redesigns the model’s architecture to do it.
Instead of predicting in parallel across thousands of experiments from many tissues at once, AlphaGenome 2 predicts DNA effects for one condition at a time, using the target biological condition as an input at every stage of its architecture. That allows it to learn how the same stretch of DNA is read differently across cell types.
The shift is the point of version 2: the same genome, read cell by cell, instead of averaged across a whole tissue.
Trained on More Than 100 Million Cells
DeepMind trained AlphaGenome 2 on both bulk and single-cell data from human and mouse samples, including more than 107 million individual cell profiles across 68 tissues. DeepMind’s summary rounds this as “more than 100 million” cell profiles.
The model predicts the same core readouts of DNA function as the original AlphaGenome — gene expression, histone modifications and chromatin accessibility — and adds DNA methylation, a widely requested readout.
Because the model separates a cell’s biological state from its outputs, it can also perform cross-modality imputation: if a scientist inputs just one measurement, such as single-cell gene activity, AlphaGenome 2 can predict unmeasured properties, such as chromatin accessibility — helping researchers decide which experiments to run next.
The model can also resolve activity inside specialised cells — such as distinct heart muscle or blood immune cells — to help uncover DNA variants that act only in specific cell types.
Nearly Twice as Many Variants Found
On the TenK10K dataset — a large single-cell map of 28 primary immune cell types — DeepMind reports that AlphaGenome 2 identified nearly twice as many causal variants as the original AlphaGenome fine-tuned on single-cell data: 30% versus 16%. For directional gene effects, the figures were 43% versus 23%.
DeepMind also reports improved cell-type specificity: on genome regions outside its training data, AlphaGenome 2 explains significantly more of how gene expression differs across cell types within a tissue — tracking continuous changes in gene expression, such as when human blood stem cells mature into red blood cells and immune lineages. Its predictions were closer to observed gene expression, DeepMind reports, in a heatmap of gene expression for maturing red blood cells.
These are DeepMind’s own benchmark figures from its launch post — no independent peer review or third-party replication had been published as of 11 November 2026, so they should be read as the company’s own claims, not as independently verified results.
The IBD Example: One Variant, One Cell Type
DeepMind illustrates the advance with a concrete case. A recent landmark study showed that a variant in a distant regulatory switch drives inflammatory bowel disease (IBD) by overactivating the ETS2 gene in gut immune cells (macrophages).
Making predictions across various immune cell types in the gut, AlphaGenome 2 reproduced this finding computationally, according to DeepMind. It singled out the causal variant and showed that it selectively boosts the ETS2 gene in gut macrophages, leaving other cell types and genes largely unaffected.
That selectivity is exactly what a broad-tissue model cannot show. A variant that matters in one cell type and not in others is invisible when tissue is averaged — and it is precisely such variants that AlphaGenome 2 is built to find.
New Cell Types and States — Without Retraining
AlphaGenome 2 can also generalise to entirely new single-cell experiments without being retrained. Scientists can upload a single-cell RNA-seq snapshot of gene activity from just a few dozen cells; that gene activity is compressed into a cell-state representation, which is then used to condition the model on unseen states.
DeepMind says the model can match or outperform models trained or fine-tuned directly on those datasets. As an example, given unseen single-cell data from a COVID-19 mRNA vaccine study, AlphaGenome 2 distinguished vaccine-activated immune T cells from resting ones, and showed 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.
A model that can read a cell state it has never seen before turns a static cell atlas into a predictive engine — DeepMind’s own description of what AlphaGenome 2 is for.
Linking Human Genetics to Single Cells
One of the hardest problems in human genetics is connecting the thousands of trait- and disease-associated variants found in genome-wide association studies (GWAS) to the cells where they actually act. AlphaGenome 2 attacks that problem in two ways, according to DeepMind.
First, it extends enhancer–gene mapping to primary human cell types — such as brain astrocytes, which support and protect neurons — helping researchers connect regulatory variants to their target genes in the specific cells where disease unfolds. Many disease-linked variants sit in distant regulatory switches called enhancers, and experimental CRISPR screens mostly use stable, lab-grown cell lines, so scientists often lack these wiring diagrams for specific cells in the human body.
Second, DeepMind has precomputed AlphaGenome 2 predictions for trait-associated variants across hundreds of human cell and tissue conditions. Across complex traits ranging from height to type 2 diabetes, the precomputed variant scores help pinpoint the specific variants and cell types most likely to be involved — and, when confident, the model also predicts whether a variant turns a gene’s activity up or down.
- What it predicts: gene expression, chromatin accessibility, histone marks and DNA methylation for a chosen cell type or state.
- What it was trained on: more than 107 million individual cell profiles (human and mouse, bulk and single-cell data) across 68 tissues.
- What it found: on TenK10K, 30% of causal variants vs 16% for the original model; 43% vs 23% for directional gene effects (DeepMind’s own figures).
- What it is not: a clinical or diagnostic tool — research use only, not validated or approved for clinical or diagnostic purposes.
Who Can Use It — and What It Is Not
AlphaGenome 2 is available from 11 November 2026 for non-commercial research via the AlphaGenome API, with interactive Colab tutorials for expression conditioning, and in private preview for commercial research. AlphaGenome 2 Skills are coming soon to Google Antigravity, DeepMind’s agentic development platform.
DeepMind is also running a community spotlight programme for interesting AlphaGenome 2 use cases, with selected projects possibly eligible for additional quota and credits; the programme closes on 30 November.
The company is working with collaborators including the Wellcome Sanger Institute to profile an even wider spectrum of human cell types, developmental lineages and perturbed cellular states. The first AlphaGenome was previewed in June 2025 and published in Nature in January 2026.
But the limit is stated plainly by DeepMind itself: AlphaGenome and AlphaGenome 2 are intended for research use only and “have not been validated or approved for clinical or diagnostic purposes”. The model cannot read a person’s genome clinically, diagnose disease, or guide treatment — it is a research tool whose predictions scientists must then test.
AlphaGenome 2 is a sharper lens on the genome, not a medical device — and its value will be proved, or not, by the discoveries researchers make with it in the months ahead.
By Akhtar Saleem Khan
Head of IT | The BNN Media – The Best News Network
News Beyond the Borders
