AlphaGenome 2 vs AlphaGenome 1: What Changed?

Paper-cut illustration in coral red, butter yellow and deep teal showing a DNA strand examined cell by cell.

Google DeepMind has a new DNA model, and this time it reads your genome cell by cell. AlphaGenome 2, announced on 11 November 2026, predicts how a change in DNA affects specific human cell types — the exact place where most diseases actually start.

What AlphaGenome 1 already did

The first AlphaGenome, previewed in June 2025 and published in Nature in January 2026, was a breakthrough of its own. Give it a DNA sequence and it predicted the molecular effects of that sequence: how strongly nearby genes would be expressed, which parts of the DNA would be accessible, and how chemical marks on the DNA would change.

The catch was resolution. AlphaGenome 1 predicted across broad tissues and cell lines — an average across many cell types at once. That average hides the detail that matters most: a DNA variant can be harmless in most cells and decisive in one rare cell type, and a tissue average dilutes that signal until it disappears.

What AlphaGenome 2 does differently

AlphaGenome 2 is a conditional model. It takes a DNA sequence of up to a million letters, plus a chosen condition: one tissue, cell type or cell state at a time. It then predicts gene expression, chromatin accessibility, histone marks and DNA methylation for that condition specifically.

A researcher can even feed in data from their own single-cell experiment — as few as a few dozen cells — and the model adapts to that condition without retraining.

DeepMind trained the model on more than 107 million individual cell profiles across 68 tissues, using human and mouse bulk and single-cell data. It is the scale of that training data that makes the cell-level predictions possible.

AlphaGenome 1 AlphaGenome 2
Prediction level Broad tissues and cell lines One cell type or state at a time
Training data Bulk tissue data 107M+ individual cell profiles, 68 tissues
Adapts to your own cells No Yes, from a few dozen cells, no retraining
Genome-wide atlas Yes, all 9 billion single-letter changes No — too expensive to compute
Access AlphaGenome API AlphaGenome API, non-commercial research

The numbers that matter

DeepMind tested both models on TenK10K, a large single-cell map of 28 primary immune cell types. The task: find the DNA variants that actually change gene expression.

AlphaGenome 2 recovered 30% of the putatively causal variants at 90% precision. AlphaGenome 1, fine-tuned on single-cell data, recovered 16%. On calling the direction of a gene's effect — whether a variant turns a gene up or down — the new model got 43% right, against 23% for the first model.

In DeepMind's words, the new model identifies nearly twice as many variants that change gene expression in cells. The paper adds a caution: even at high precision, it still recovers only a minority of the variants statistical analyses nominate as likely causal. Thirty percent is a big step from sixteen — and still a long way from all of them.

What it can and cannot do

AlphaGenome 2 is available today for non-commercial research through the AlphaGenome API, with commercial research in private preview.

It does not replace its predecessor. AlphaGenome 2 predictions cost more compute, so a genome-wide atlas like the one built with AlphaGenome 1 is impractical with it. AlphaGenome 1 stays in service for broad, cheap scans; AlphaGenome 2 is for focused, cell-specific questions.

And it is not a medical tool. DeepMind states plainly that both AlphaGenome models are for research use only and have not been validated or approved for clinical or diagnostic purposes. This is not a tool that reads your personal genome — it is a research instrument that helps scientists find where disease begins, cell by cell.

Key takeaways

  • DeepMind announced AlphaGenome 2 on 11 November 2026; it predicts DNA variant effects in specific cell types and states.
  • It was trained on more than 107 million individual cell profiles across 68 tissues.
  • On the TenK10K immune benchmark it found 30% of causal variants, against 16% for AlphaGenome 1.
  • It can adapt to a researcher's own single-cell data from a few dozen cells, without retraining.
  • It is available now for non-commercial research via the AlphaGenome API; commercial access is in private preview.
  • It is for research only — not validated or approved for clinical or diagnostic use.

Conclusion

AlphaGenome 2 is not a bigger AlphaGenome. It is a more precise one: the same DNA, read one cell type at a time, with nearly twice as many disease-linked variants found on the key benchmark. The gap between 30% and 100% is a reminder that this is still early science — but the search just got a lot narrower.

FAQ

What is the difference between AlphaGenome 2 and AlphaGenome 1?

AlphaGenome 1 predicts DNA effects across broad tissues. AlphaGenome 2 predicts them for one specific cell type or state at a time, and can adapt to a researcher's own single-cell data.

How much better is AlphaGenome 2?

On the TenK10K benchmark of 28 immune cell types, it recovered 30% of causal expression-linked variants at 90% precision, against 16% for the first model — nearly twice as many.

Can I use AlphaGenome 2 for medical diagnosis?

No. DeepMind says both AlphaGenome models are for research use only and have not been validated or approved for clinical or diagnostic purposes.

Sources

Related reading on Daily Tech & AI:

Researched with AI assistance and reviewed for accuracy.

Follow on WhatsApp

Follow Daily Tech & AI on WhatsApp for the latest tech and AI updates: https://www.whatsapp.com/channel/0029Vb9XzdHDJ6H4TFNAl61r

Comments