Datasets:

maomlab
/

ChAFF

license: cc0-1.0
language:
  - en
tags:
  - bioassay
pretty_name: CHAFF
size_categories:
  - 10K<n<100K
dataset_info:
  - config_name: CHAFF
    features:
      - name: Type
        dtype: string
      - name: AID
        dtype: int64
      - name: CID
        dtype: int64
      - name: SMILES
        dtype: string
    splits:
      - name: train
        num_bytes: 5834311
        num_examples: 69777
    download_size: 1913364
    dataset_size: 5834311
  - config_name: ChAFF
    features:
      - name: Type
        dtype: string
      - name: AID
        dtype: int64
      - name: CID
        dtype: int64
      - name: SMILES
        dtype: string
    splits:
      - name: train
        num_bytes: 5970580
        num_examples: 71400
    download_size: 1963227
    dataset_size: 5970580
configs:
  - config_name: CHAFF
    data_files:
      - split: train
        path: CHAFF/train-*
  - config_name: ChAFF
    data_files:
      - split: train
        path: ChAFF/train-*

Dataset description

This dataset collection contains ~70K curated active compound lists from 73 different PubChem BioAssay (AID) datasets, focusing on known assay interference artifacts. We downloaded raw assay results from PubChem using their AID identifiers and extracted only the compounds labeled as "Active."

Then We applied SMILES standardization using RDKit and MolVS, including molecule sanitization and fragment removal.

Each dataset includes the following columns:

Type: Interference type (e.g., Autofluorescence, REDOX)
AID: PubChem Assay ID
CID: PubChem Compound ID (Some CID would not match its SMILES as we did SMILES sanitization)
SMILES: Curated chemical representation

The final dataset is suitable for training and evaluating machine learning models.

If you are looking for a large combined dataset with various AIDs, click here.
If you prefer individual datasets for each AID, click here.
Current version of our repository: 1.0.0

List of PubChem AIDs included:

632, 1641, 1730, 1857, 1926, 435026, 504689, 720541, 1159604, 587, 588, 589, 590, 591, 592, 593, 594, 709, 923, 1480, 1483, 1696, 1775, 1776, 2124, 2757, 588517, 588620, 624483, 720675, 720678, 720680, 720681, 720682, 720686, 720687, 584, 585, 1476, 1478, 485294, 485341, 411, 1006, 1269, 1379, 1891, 2515, 2530, 366887, 366889, 366891, 488838, 493175, 588342, 588498, 602357, 602358, 602364, 602474, 602475, 602476, 602477, 624030, 652016, 720522, 720835, 1224835, 1347047, 672, 682, 936, 878, 888, 929, 1234

Dataset processing

If you are interested in our dataset curation process, follow these scripts in our repository.

st1_download_pubchem.py

Download bioassay datasets from PubChem using a single AID (Assay ID) and save them in CSV format.

st2_run_download_pubchem.py

Automate the download process for multiple AIDs by allowing you to input a list of AIDs, sequentially downloading each dataset.

st3_extract_active_compounds.py

Parse each dataset and filters rows labeled as "Active".

st4_smiles_curation.py

Standardize and validate 'CanonicalSMILES' column. Apply sanitization, standardization, and fragment removal using RDKit and MolVS.

st5_detergent_smiles_curation.py

Special handling for AIDs 585, 584, 1476, 1478, 485341, 485294 These datasets are processed to remove overlapping compounds based on detergent-related assay pairs. (Without detergent - With detergent). This ensures non-specific binders (likely aggregators) are excluded.

AID 585 → removed 48 compounds also active in AID 584
AID 1476 → removed 439 compounds also active in AID 1478
AID 485341 → removed 44 compounds also active in AID 485294