QMAP benchmark

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.

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

Method	Year	e. coli min PCC	e. coli mean	e. coli max PCC	Source Code
Linear model on ESM2 embeddings	N/A	0.32.	0.36	0.41	eval_prev_works/Linear
J. Witten and Z. Witten	2019	0.47	0.51	0.56	eval_prev_works/Antimicrobial-Peptides
J. Cai et al	2025	0.47	0.52	0.56	eval_prev_works/AMP_regression_EC_SA

High efficiency - e. coli

Method	Year	e. coli min PCC	e. coli mean	e. coli max PCC	Source Code
Linear model on ESM2 embeddings	N/A	0.06	0.16	0.22	eval_prev_works/Linear
J. Witten and Z. Witten	2019	0.10	0.22	0.33	eval_prev_works/Antimicrobial-Peptides
J. Cai et al	2025	0.20	0.29	0.33	eval_prev_works/AMP_regression_EC_SA

Full - hc50

Method	Year	e. coli min PCC	e. coli mean	e. coli max PCC	Source Code
Linear model on ESM2 embeddings	N/A	-0.18	0.07	0.29	eval_prev_works/HemoLinear

Install QMAP-benchmark

pip install qmap-benchmark

Documentation

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

Reproduce the results or the paper

To reproduce the results of the paper, you must first fetch and prepare the data.

Then, you can run what you are interested in:)

The repo is structured as follow:

• data: Code to download and prepare the data used in the project.

• eval_prev_works: Code to evaluate previous methods on the QMAP benchmark and to draw the figures showing the performances.

• figures: Contains multiple notebooks to generate additional visualizations.

• QMAP: The PyPi package code.

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}
}