Google DeepMind Launches AlphaGenome 2 — DNA Model Doubles Genetic Variant Detection Cell by Cell

Google DeepMind released AlphaGenome 2 today, a DNA analysis model that nearly doubles its predecessor's ability to identify which genetic variants actually cause disease — and it does it cell by cell, not just genome-wide. That specificity matters: most of the genome's functional effects depend on what kind of cell you're in.

Genetic variants are differences in DNA sequence between individuals. Most are harmless. A small fraction alter how genes are expressed — turned on or off, up or down — in ways that drive disease. Pinpointing which variants are causal, and in which tissues, has been one of the core bottlenecks in translating genomic research into treatments. Prior models predicted variant effects using the genome sequence alone, treating all cell types the same.

What's New in AlphaGenome 2

AlphaGenome 2 takes a fundamentally different input structure. Rather than analyzing a DNA sequence in isolation, the model accepts both a DNA sequence and a specified cell type as joint inputs — and predicts how a given variant will alter gene expression specifically within that cell context. Training data covered more than 100 million individual cell profiles, giving the model broad coverage across tissue types.

The performance improvement is significant:

• 30% causal variant detection rate in benchmark testing
• Versus 16% for the original AlphaGenome — a near-doubling
• The benchmark tests the model's ability to identify which variant among a set of candidates is functionally responsible for a disease association

Access and Availability

As of today, November 12, 2026, AlphaGenome 2 is available for non-commercial research use. Commercial access is in private preview; organizations interested in clinical or drug-discovery applications can request access directly from DeepMind.

The non-commercial release follows AlphaFold's open-access model, which became a standard reference tool for protein structure prediction within two years of its public release. DeepMind has not yet announced commercial pricing tiers.

Why It Matters for Drug Discovery and Clinical Genomics

For drug discovery, identifying the causal variant in a disease-linked genomic region is often the first step toward understanding mechanism — and mechanism drives target selection. A model that doubles detection accuracy could meaningfully compress early-stage research timelines for programs rooted in genetic evidence.

For clinical genomics, cell-type specificity addresses a long-standing problem: variants with identical DNA sequences behave differently in liver cells versus neurons versus immune cells. Diagnoses that depend on tissue-specific effects — certain inherited cancers, neurological disorders, rare diseases — stand to benefit most.

For the field broadly, AlphaGenome 2 adds to a growing suite of DeepMind biology tools — AlphaFold 3, AlphaMissense, and now AlphaGenome 2 — that together cover structure, pathogenicity, and regulatory function. The convergence suggests DeepMind is building toward integrated end-to-end genetic analysis rather than point solutions.

What to Watch

The 30% detection rate is a substantial improvement but not a solved problem — 70% of causal variants are still missed. Watch for peer-reviewed validation of benchmark claims and early adoption reports from large-scale genomics consortia. Commercial pricing terms, when announced, will determine whether this tool reaches clinical labs or stays confined to well-funded research institutions.