QMAP benchmark and toolkit

A domain-specific homology aware benchmark that ensure robust performance evaluation and comparison between models predicting MIC and/or HC50 (regression).

Features:

• Regression on consensus bacterial MIC and mammal HC50
• Predefined splits, to ensure comparability
• N-terminal acetylation and C-terminal amidation
• Common noncannonical amino acids O: Ornithine and B: 2,4-Diaminobutyric acid
• Non-cannonical peptides as SMILES
• Intrachain bonds:
• Machine learning ready dataset
• Extendable training dataset, so you can use your own data.

Install QMAP-benchmark

pip install qmap-benchmark

Get started

# Import the DBAASP training dataset and the benchmark
from qmap import QMAPBenchmark, DBAASPDataset, train_test_split

dataset = DBAASPDataset()
# We need to test on each split, and train on an independent set for each split.
for split in range(5):
    benchmark = QMAPBenchmark(split=split)

    train_mask = benchmark.get_train_mask(dataset.sequences)
    train_dataset = dataset[train_mask]
    train_dataset, val_dataset = train_test_split(train_dataset, test_size=0.15, random_state=42)
    
    # Train on train_dataset ...
    ...
    # Predict the benchmark dataset
    preds = ...
    
    # The benchmark automatically computes the metrics!
    test_perf = benchmark.compute_metrics(preds)
    print(test_perf)

Leaderboard

Add an entry:

Here is the leaderboard. Please open an issue to add your model to the leaderboard with the code to reproduce the results, and the reference paper.

Full - e. coli

High efficiency - e. coli

Full - hc50

Documentation

The documentation formatted as markdown is available in QMAP/docs/references
Examples are shown in QMAP/docs/examples

Please Cite

Please cite us if you find yourself using our work

@misc{lavertu_qmap_2026,
    title = {{QMAP}: {A} {Benchmark} for {Standardized} {Evaluation} of {Antimicrobial} {Peptide} {MIC} and {Hemolytic} {Activity} {Regression}},
    url = {https://www.biorxiv.org/content/10.64898/2026.02.03.703041v1},
    doi = {10.64898/2026.02.03.703041},
    publisher = {bioRxiv},
    author = {Lavertu, Anthony and Corbeil, Jacques and Germain, Pascal},
    month = feb,
    year = {2026}
}