AlphaGenome
GoogleAlphaGenome is Google DeepMind's genomic sequence model for predicting gene expression, splicing, chromatin features, contact maps, and variant effects. It accepts sequences up to one million base pairs and produces most outputs at single-base-pair resolution through an API, a precomputed Atlas, and a gated downloadable checkpoint.
The repository documents both API and local-checkpoint surfaces; the downloadable model remains subject to separate non-commercial terms.
Openness
2 high confidence- weights
- open(gated downloadable checkpoint files)
- data
- open(training TFRecords for all 12 predicted modalities published in a public, unauthenticated Google Cloud Storage bucket (alphagenome-datasets) that the alphagenome_research data loader reads from)
- code
- partial(Apache-2.0 API client, examples, and notebook
- license
- AlphaGenome-Model-Terms(non-commercial use only)
A separate Google DeepMind repository publishes the model's JAX implementation with a loader for its training files, and those files sit in a public Google Cloud Storage bucket that anyone can list and download without signing in. The downloadable checkpoint is still covered by the AlphaGenome Model Terms, which forbid commercial use.
- https://deepmind.google.com/science/alphagenome/model-terms recorded 2026-10-21
AlphaGenome model terms govern the downloadable model and limit the grant to non-commercial use.
- https://huggingface.co/api/models/google/alphagenome-all-folds recorded 2026-10-21
gated: auto; private: false; license_name: alphagenome; checkpoint metadata, manifests, and weight shards present; gate says non-commercial use only.
- https://raw.githubusercontent.com/google-deepmind/alphagenome/main/README.md recorded 2026-10-14
Repository supplies an Apache-2.0 API client and examples, describes hosted predictions and Atlas, and publishes no pretraining code or corpus.
- https://storage.googleapis.com/storage/v1/b/alphagenome-datasets/o?delimiter=/&prefix=v1/train/ALL_FOLDS/HOMO_SAPIENS/TRAIN/&maxResults=100 recorded 2026-10-21
Unauthenticated GCS JSON API lists prefixes ATAC/, CAGE/, CHIP_HISTONE/, CHIP_TF/, CONTACT_MAPS/, DNASE/, PROCAP/, RNA_SEQ/, SPLICE_JUNCTIONS/, SPLICE_SITES/, SPLICE_SITE_USAGE/ under v1/train/ALL_FOLDS/HOMO_SAPIENS/TRAIN/, with real object entries (size, md5Hash, mediaLink) for the training TFRecord shards.
Adoption
2 medium confidenceThe gated Hugging Face endpoint reports zero downloads, so GitHub stars are used as the adoption signal instead.
- https://api.github.com/repos/google-deepmind/alphagenome recorded 2026-09-10
stargazers_count: 2080; archived: false; fork: false.
Capability
2 high confidenceAlphaGenome's strength is its resolution span: sequences up to one million bases are resolved to single-base outputs. Its only input is a DNA sequence, so it does not also count as accepting multiple sensors, and no adaptation or generative pathway is documented beyond a broad prediction surface.
- https://raw.githubusercontent.com/google-deepmind/alphagenome/main/README.md recorded 2026-09-10
Documents gene-expression, splicing, chromatin, and contact-map output modalities over sequences up to one million bases, mostly at single-base resolution.
Verified 2026-09-10