AlphaGenome 2 Reads DNA One Cell Type at a Time. What It Can and Cannot Do

AlphaGenome 2 Reads DNA One Cell Type at a Time. What It Can and Cannot Do

Published
AuthorDevansh

On November 11, 2026, Google DeepMind announced AlphaGenome 2, a model that predicts how DNA sequences and their variations affect specific human cell types. The first AlphaGenome, published in Nature in January 2026, read DNA across thousands of tissue samples at once, giving an average answer. AlphaGenome 2 changes the question: give it a stretch of DNA plus a specific cell type, cell state, or condition, and it predicts how that stretch is read in that cell, which genes are active, how the DNA is packaged and accessed, and how a single-letter change would alter any of it. That matters because a variant can matter only in one specialised cell, or only when a cell is activated by disease, and its signal disappears when you average across a whole tissue.

What changed from the first model

AlphaGenome 2 is a conditional model: it takes a specific biological condition alongside the input DNA sequence and supplies that condition throughout the network, so the same DNA segment is read differently across cell types. It was trained on more than 100 million individual cell profiles across 68 tissues, from human and mouse bulk and single-cell data, turning static cell atlases into a predictive engine for human genetics. It can also adapt to a condition generated from a researcher's own single-cell experiment. On DeepMind's flagship test, TenK10K, a single-cell map of 28 primary immune cell types, AlphaGenome 2 recovered 30 percent of putatively causal expression-linked variants at 90 percent precision, compared with 16 percent for AlphaGenome 1 fine-tuned on single-cell data, and recovered the direction of the gene effect for 43 percent of variants, against 23 percent before. It also reproduced a known inflammatory bowel disease mechanism involving ETS2. Those numbers are DeepMind's own, from a preprint that had not been peer-reviewed at launch, on an immune-cell dataset, so performance in other parts of the body is less established.

What it cannot do

The limits are stated plainly by DeepMind and they are the point of the post. AlphaGenome 2 still misses most of what it is looking for: even at high precision, it recovers only a minority of the expression-linked and disease-associated variants that statistical analyses nominate as likely causal. It covers the core prediction, not a diagnosis. DeepMind states that AlphaGenome models are intended for research use only and have not been validated or approved for clinical or diagnostic purposes. The model itself cannot be downloaded and run locally. Treat any claim that an AI model can diagnose you from your DNA as a red flag, no matter how impressive the benchmark. The honest sentence is narrower and still remarkable: a model can now predict, cell type by cell type, which DNA changes are likely to matter, and researchers can test those predictions in the lab instead of guessing across a whole tissue.

Who can use it, and how

AlphaGenome 2 is available now for non-commercial research via the AlphaGenome API, with interactive Colab tutorials for expression conditioning, and in private preview for commercial research. DeepMind also plans AlphaGenome 2 Skills in Google Antigravity, its agentic development platform. For builders, the pattern is the transferable part: a general model, conditioned on the specific context that changes the answer, beats a bigger average. The same idea applies to agents and infrastructure: a router that knows the workload, a monitor that knows what healthy means for this service, and a model that is told the exact condition it is deciding under. If you run models or agents on your own hardware and want the serving layer right, our vLLM serving guide covers the measure-first discipline, and our Ollama guide covers running a model on your own machine. If you want the infrastructure around an AI workload checked, book a free infra audit with DevOpsBY.

Disclaimer: This post is educational only, not financial or medical advice. Figures are as reported by the sources on 2026-11-13 and may change - always confirm version numbers, requirements, and scope in the official docs and paper before acting on them.

Related Articles

DevOps & Infrastructure Services

Need hands-on help? We build and manage the infrastructure behind the concepts covered in this article.

DevOpsBy Assistant

How can we help?

Select a topic to get started: