MoleculeACE Dataset

Overview

The MoleculeACE (Molecule Activity Cliff Estimation) dataset contains bioactivity data for 30 different ChEMBL targets with corresponding protein sequences. This dataset is designed to assess how well molecular machine learning models can handle activity cliffs - pairs of molecules that are structurally similar but show large differences in biological activity.

Dataset Structure

The dataset is organized with each ChEMBL target as a separate configuration:

from datasets import load_dataset

# Load a specific task
dataset = load_dataset("your-username/moleculeace", "CHEMBL204_Ki")

# Access train/test splits
train_data = dataset["train"]
test_data = dataset["test"]

Features

Each sample contains:

• smiles: SMILES representation of the molecule
• activity_value: Activity value (pEC50/pKi)
• protein_sequence: Amino acid sequence of the target protein
• chembl_id: ChEMBL target identifier
• task: Task name

Available Tasks

• CHEMBL1862_Ki
• CHEMBL1871_Ki
• CHEMBL2034_Ki
• CHEMBL2047_EC50
• CHEMBL204_Ki
• CHEMBL2147_Ki
• CHEMBL214_Ki
• CHEMBL218_EC50
• CHEMBL219_Ki
• CHEMBL228_Ki
• CHEMBL231_Ki
• CHEMBL233_Ki
• CHEMBL234_Ki
• CHEMBL235_EC50
• CHEMBL236_Ki
• CHEMBL237_EC50
• CHEMBL237_Ki
• CHEMBL238_Ki
• CHEMBL239_EC50
• CHEMBL244_Ki
• CHEMBL262_Ki
• CHEMBL264_Ki
• CHEMBL2835_Ki
• CHEMBL287_Ki
• CHEMBL2971_Ki
• CHEMBL3979_EC50
• CHEMBL4005_Ki
• CHEMBL4203_Ki
• CHEMBL4616_EC50
• CHEMBL4792_Ki

Usage Example

from datasets import load_dataset

chembl204_data = load_dataset("your-username/moleculeace", "CHEMBL204_Ki")
train_samples = chembl204_data["train"]

# Access features
for sample in train_samples:
    smiles = sample["smiles"]
    activity = sample["activity_value"]
    protein_seq = sample["protein_sequence"]
    # ... process your data

Dataset Statistics

• Total tasks: 30 ChEMBL targets
• Data splits: Train/Test for each task
• Features: SMILES, activity values, protein sequences, ChEMBL IDs
• Domain: Molecular bioactivity prediction
• Focus: Activity cliff estimation

Citation

This dataset was introduced in:

@article{vanTilborg2022,
    title={Exposing the Limitations of Molecular Machine Learning with Activity Cliffs},
    author={van Tilborg, Derek and Alenicheva, Alisa and Grisoni, Francesca},
    journal={Journal of Chemical Information and Modeling},
    year={2022},
    publisher={ACS Publications}
}

Data Source

• Original MoleculeACE data: GitHub Repository
• Protein sequences: ChEMBL Database
• Reference: https://pubs.acs.org/doi/10.1021/acs.jcim.2c01073

License

Please refer to the original MoleculeACE repository for licensing information.