README.md

metadata

pretty_name: LULA Data Public Train/Test
license: other
tags:
  - bioactivity
  - chemistry
  - protein-ligand-binding
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train/*.parquet
      - split: test
        path: data/test/*.parquet

LULA Data Public Train/Test

This dataset contains 6,819,817 protein-small-molecule pairs with binary bioactivity labels and fixed train/test splits.

Fields

• pair_id: stable identifier for the protein-molecule pair.
• canonical_smiles: canonical molecule SMILES.
• protein_sequence: amino-acid sequence.
• uniprot_id: normalized UniProt identifier.
• label: 1 for binder/active and 0 for non-binder/inactive.
• split: train or test.
• sources: public databases contributing evidence for the pair.
• evidence_count: number of contributing source records.

Splits

Split	Rows	Non-binder (0)	Binder (1)
train	6,137,835	1,321,443	4,816,392
test	681,982	179,488	502,494

Provenance

The dataset aggregates publicly available bioactivity records from BindingDB, ChEMBL, GLASS2, GtoPdb, NPASS, Papyrus, and TTD. The sources field records which databases contributed evidence to each pair.

License and attribution

This dataset uses the Hugging Face other license designation because the contributing databases retain their own licenses and terms. Please cite the source databases named in each row's sources field:

• BindingDB
• ChEMBL
• GLASS2
• IUPHAR/BPS Guide to PHARMACOLOGY
• NPASS
• Papyrus
• Therapeutic Target Database

The original source records remain subject to the licenses and terms of their source databases. Om's normalization and aggregation do not replace those source-specific terms.

Label interpretation

0 indicates an inactive or non-binding classification in the contributing source data. It does not mean binding is impossible under every experimental condition.