alphagenome_dataset / README.md
Zhongning's picture
Upload folder using huggingface_hub
243963d verified
|
Raw
History Blame Contribute Delete
2.85 kB
metadata
license: other
tags:
  - GRCh38
  - GRCm38
language:
  - en
  - zh

AlphaGenome Dataset

Dataset Description

AlphaGenome Dataset is a curated package of reference genomes, annotation files, and TFRecord data for use with the OneScience AlphaGenome model.

The reference resources include human GRCh38.p13, mouse GRCm38.p6, FAI indexes, Gencode annotations, splice-site annotations, and polyA annotations. The TFRecord data contains 12 bundles from the HOMO_SAPIENS VALID split, with 25 gzip-compressed shards in each bundle.

Supported Tasks

This dataset has been adapted for OneScience-Group/alphagenome/. By default, the model repository expects this dataset to be located at data/alphagenome_dataset within the model package. It is used for preflight checks, interval inference, variant scoring, and track-prediction evaluation.

Dataset Format and Structure

TFRecord file paths follow this pattern:

v1/train/ALL_FOLDS/HOMO_SAPIENS/VALID/<BUNDLE>/data_chrAll_01-25.gz.tfrecord

The bundles comprise ATAC, DNASE, PROCAP, CAGE, RNA_SEQ, CHIP_TF, CHIP_HISTONE, CONTACT_MAPS, SPLICE_SITES, SPLICE_SITE_USAGE, SPLICE_JUNCTIONS, and SPLICE_SITE_POSITIONS. Each sample contains a DNA sequence, interval chromosome/start/end metadata, a target tensor, and a mask; see metadata/schema.yaml for dtype and resolution.

How to Use the Dataset

Files and Download

Standalone download:

hf download --dataset OneScience-Group/alphagenome_dataset --local-dir .

Use within the model package:

hf download --dataset OneScience-Group/alphagenome_dataset --local-dir data/alphagenome_dataset

Data Placement and Validation

When used as a model dependency, the following directory structure is recommended:

<model-package>/data/alphagenome_dataset/
├── manifest.yaml
├── metadata/
├── reference/
└── v1/train/ALL_FOLDS/HOMO_SAPIENS/VALID/

Quickly validate file presence, sizes, and gzip readability:

python scripts/validate_dataset.py --root .

Add --check-sha256 to perform a complete SHA256 validation.

Official OneScience Information

Limitations and License

license: mixed license_note: > AlphaGenome training, validation and test datasets contain reprocessed data from FANTOM5, 4D Nucleome, ENCODE and GENCODE. Usage is subject to the respective source-specific licenses and data-use policies. Other unspecified materials are licensed under CC BY-NC 4.0.