MONDAY, NOVEMBER 16, 2026 Tech news, minus the noise.
N News Cast
LATEST
Home  /  AI
AI

DeepMind's AlphaGenome 2 Reads DNA One Cell at a Time

Google DeepMind's AlphaGenome 2 predicts how DNA variants affect gene regulation in individual cell types, trained on more than 100 million single-cell profiles.

DeepMind's AlphaGenome 2 Reads DNA One Cell at a Time
Source: National Institute of Standards and Technology (Public domain) via Wikimedia Commons

AlphaGenome 2 predicts variant effects cell by cell

Google DeepMind announced AlphaGenome 2 on November 11, 2026, a deep learning model that predicts how DNA variants affect gene regulation in individual cell types and states. It was trained on more than 100 million individual cell profiles, and it is available now for non-commercial research through the AlphaGenome API, with commercial research in private preview.

The problem it attacks is a real blind spot in genomic modeling. A DNA variant can matter only in one specialized cell type, or only in a cell switched on by disease, and its signal gets diluted when you average across a whole tissue. Earlier models predicted molecular activity across broad tissues. AlphaGenome 2 instead conditions each prediction on a selected tissue, cell type, or cell state, including conditions generated from a researcher's own single-cell experiment. The same stretch of DNA gets read differently depending on the cell, and the model tries to capture that.

What AlphaGenome 2 does better than version one

The first AlphaGenome generated fixed outputs for bulk tissues and cell lines at once from a one-million-letter DNA sequence. AlphaGenome 2 evaluates one condition at a time and supplies that condition throughout the network. That sounds like a small architectural choice, but it is the whole point: cell identity is not an output label, it is an input that shapes every prediction.

DeepMind's paper puts numbers on the improvement. On TenK10K, a single-cell immune dataset spanning 28 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. Nearly doubling the recovery rate at the same precision is a meaningful jump for researchers hunting the variants behind disease.

The paper is honest about the limits too. Even at high precision, the model recovers only a minority of the expression-linked and disease-associated variants that statistical analyses nominate as likely causal. It is a better flashlight, not a finished map.

How researchers can actually use it

At launch, AlphaGenome 2 is available through the AlphaGenome API for non-commercial research, with expression-conditioning Colab tutorials to get started. DeepMind says non-commercial API use is free, with query rates varying with demand. Commercial research sits in private preview, and DeepMind also plans AlphaGenome 2 Skills for Google Antigravity, its agentic development platform.

DeepMind precomputed two resources for researchers who do not want to run large-scale comparisons themselves. That matters because the model is not light: training on more than 100 million cell profiles across human and mouse bulk and single-cell data puts it out of reach for most labs to run from scratch. The API and the precomputed data are the practical on-ramps.

Genomics Limited CEO Peter Donnelly said he looked forward to incorporating AlphaGenome 2 into the company's Mystra platform, an early sign that the model will show up in commercial genomics pipelines once the private preview opens wider.

The context: AI models are getting narrower and sharper

AlphaGenome 2 arrives in a month crowded with frontier releases. OpenAI shipped GPT-6 Cyber for approved security teams, Moonshot put the 2.8-trillion-parameter Kimi K3.1 Pro weights on Hugging Face, and Anthropic split its new Claude 5.2 line into Fable and Mythos access paths. The pattern across all of them is the same: the general-purpose race continues, but the most interesting releases are specialized models aimed at one hard problem.

Genomics is a good test of that bet. The first AlphaGenome and the AlphaGenome Atlas, which offered predictions for all 9 billion possible single-letter changes in the human genome, showed that AI could scale across the genome. AlphaGenome 2 tries to go deep instead of wide, and the TenK10K numbers suggest depth is paying off.

One boundary worth stating plainly: DeepMind restricts both AlphaGenome versions to research use. Neither has been validated or approved for clinical or diagnostic purposes. If you are a patient or a clinician, this is not a tool for you yet. If you are a researcher staring at a variant list and wondering which ones matter in which cells, it is worth a look this week. Gemini 4 Argon showed the same pattern on the general-purpose side: the model matters less than the job it is aimed at.

AS
News Cast Editorial Team

We cover AI, gadgets, startups and security — original reporting and explainers, written by humans, for humans. Got a tip? Reach us via the contact page.