README.md

metadata

tags:
  - protein structure prediction
  - MMseqs
  - Jackhmmer
language:
  - en
  - zh
license: other

AlphaFold3 Dataset

Dataset Description

AlphaFold3_dataset is the companion dataset for the OneScience-Group/AlphaFold3/ model. It includes inference input examples, public search databases, MMseqs databases, sharded Jackhmmer databases, and alignment-related directories. The dataset ID is fixed as OneScience-Group/AlphaFold3_dataset.

Supported Tasks

This dataset supports the AlphaFold3 biomolecular structure prediction workflow. It provides both AlphaFold3 JSON examples that already contain MSAs and the database directories required to search for MSAs from sequences. The model can use these data for preflight checks, inference with MSAs, the MMseqs data pipeline, and the Jackhmmer data pipeline.

Dataset Format and Structure

Path	Type	Purpose	Required	Capability	Placement After Download	Notes
manifest.yaml	Manifest file	Machine-readable dataset description	Yes	All capabilities	Dataset package root directory	The LLM must parse this file
onescience_run_manifest.yaml	Runtime Manifest	Web interface read entry point	Yes	All capabilities	Dataset package root directory	Consistent with manifest.yaml
metadata/file_manifest.yaml	Integrity manifest	Records the source path, organized path, size, and SHA256 of key files	Yes	Preflight check	metadata/	Verifies consistency after organization
scripts/validate_dataset.py	Validation script	Checks directories, JSON schema, sizes, and SHA256	Yes	Data read validation	scripts/	Does not run the model
data/infer_input_data/	Input examples	AlphaFold3 JSON examples containing all_data and pressure_data	Yes	Inference and preflight checks	data/infer_input_data/	The smallest example is all_data/7r6r_data.json
data/public_databases/	Public databases	Databases for Jackhmmer and template searches	Yes	Data pipeline	data/public_databases/	The upload view uses pdb_2022_09_28_mmcif_files.tar instead of the extracted mmcif_files/ directory to avoid file-count limits
data/mmseqsDB/	MMseqs databases	MMseqs search databases	Yes	MMseqs data pipeline	data/mmseqsDB/	Includes small_bfd, mgnify, uniprot, and uniref90
data/jackhmmer_split/	Sharded database	Sharded FASTA files for Jackhmmer searches	Yes	Jackhmmer data pipeline	data/jackhmmer_split/	See the directory filenames for the shard count
data/alignDB/	Alignment database	Alignment-related data	No	Advanced workflows	data/alignDB/	Optional

How to Use the Dataset

Files and Download

Download the dataset:

hf download --dataset OneScience-Group/AlphaFold3_dataset

Download the compatible model:

hf download --model OneScience-Group/AlphaFold3/

If you use --cache_dir, you must switch to the root directory of the downloaded dataset package before validation.

Data Placement

After downloading the dataset separately, you can run validation from the dataset package root directory. To use it with the model, place this repository's data subdirectory in the model package:

<model-package>/data/alphafold3_dataset

You can do this by copying or linking the directory:

mkdir -p /path/to/model-package/data
ln -s /path/to/AlphaFold3_dataset/data /path/to/model-package/data/alphafold3_dataset

Data Read Validation

Run the following command from the dataset package root directory:

python scripts/validate_dataset.py --root .

The validation script parses metadata/file_manifest.yaml and checks that key directories exist, JSON examples are readable, and filenames, sizes, and SHA256 values are consistent. On success, it outputs dataset_validation_ok: true.

Official OneScience Information

Platform	OneScience Main Repository	Skills Repository
Gitee	https://gitee.com/onescience-ai/onescience	https://gitee.com/onescience-ai/oneskills
GitHub	https://github.com/onescience-ai/OneScience	https://github.com/onescience-ai/oneskills

Limitations and License

mmseqs is a locally generated database. The public databases and AlphaFold3-related data are subject to their respective upstream licenses and terms of use.