AlphaGenome

DNA sequence model from Google DeepMind predicting regulatory and non-coding variant effects on gene expression, splicing, and chromatin state.

Unsigned

Overview

AlphaGenome is a DNA sequence foundation model from Google DeepMind that predicts the effects of genetic variants on gene expression, splicing, chromatin accessibility, transcription-factor binding, and 3D contact maps, from sequence context up to 1 Mb, at near single-base-pair resolution. It is peer-reviewed in Nature ("Advancing regulatory variant effect prediction with AlphaGenome", 2025/2026, PMC12851941), and DeepMind has published the AlphaGenome Atlas, precomputing regulatory-impact predictions for essentially all ~9 billion possible single-nucleotide variants in the human genome -- a direct regulatory/non-coding variant-effect-prediction use case. A companion repository (`google-deepmind/alphagenome_research`) publishes the research code, scoring implementations, and evaluation datasets behind the Nature paper; this entry catalogues the actively maintained API client repository as the primary access point.

**Licence note, confirmed directly against the repository, not derived metadata**: the client code in this repository is Apache-2.0 (confirmed against the repository's own "License and Disclaimer" section and raw `LICENSE` file). Documentation and examples are CC-BY-4.0. However, the model weights and the API's own outputs (including the AlphaGenome Atlas) are governed by a separate, non-commercial AlphaGenome Terms of Use/model terms: outputs are restricted to non-commercial use, may not be used to train other ML models, and are explicitly not for clinical decision-making. Because that restriction governs the actual model and its predictions -- the genomic-AI artefact this registry catalogues -- rather than merely the client library, `license` here is recorded as `other` with `license_url` pointing to those terms, per this registry's own precedent for a model whose weights/outputs carry a real usage restriction beyond its client code's licence (`data/models/nucleotide-transformer.json`).

**Registry note on `model_card_uri`**: this points to the Hugging Face mirror of one released checkpoint (`google/alphagenome-all-folds`), which carries a standard Hugging Face model card, rather than a GA4GH Genomic AI Model Card v0.6.0-formatted document (no model has yet published one, since it is a newly drafted specification). The primary source and API client remain `github.com/google-deepmind/alphagenome`.

Details

Licence: other (full text)
Version: AlphaGenome (2025/2026 Nature release)
Category: Variant effect prediction
Homepage: https://www.alphagenomedocs.com
Repository: https://github.com/google-deepmind/alphagenome
Maintainers: Susheel Varma (@susheel, GA4GH AI Workstream / Sage Bionetworks)
Keywords: dnaregulatory-variantvariant-effectgene-expressionsplicingchromatin-accessibilitygenomics

Safety classification

No safety classification has been submitted for this entry.

Model Card

View the full GA4GH Genomic AI Model Card ↗

Hosting

github

Access

https://github.com/google-deepmind/alphagenome

Install

pip install ./alphagenome  # after: git clone https://github.com/google-deepmind/alphagenome.git

Mirrored data last synced from the Model Card at 2026-09-08T00:00:00.000Z. See the linked card for the authoritative, live version.

Provenance

Created: 2026-09-08T00:00:00.000Z
Updated: 2026-09-08T00:00:00.000Z
Last verified: 2026-09-08T00:00:00.000Z