Molecular fingerprints are strong models for peptide function prediction

Abstract Motivation Understanding peptide properties is often assumed to require modeling long-range molecular interactions, motivating complex graph neural networks and pretrained transformers. Whether such long-range dependencies are essential remains unclear. We investigate if simple, domain-specific molecular fingerprints can capture peptide function without these assumptions. Atomic-level representations aim to provide richer information than purely sequence-based models and better efficiency than structural ones. Results Across 132 datasets, including LRGB and five additional peptide benchmarks, models using count-based ECFP, Topological Torsion, and RDKit fingerprints with LightGBM achieve state-of-the-art accuracy. Despite encoding only short-range molecular features, these models outperform GNNs and transformer-based approaches. Control experiments confirm that fingerprints, though inherently local, suffice for robust peptide property prediction. Our results challenge the presumed necessity of long-range interaction modeling and highlight molecular fingerprints as efficient, interpretable, and lightweight alternatives. Supplementary information All code and data are available on GitHub and Zenodo: https://github.com/scikit-fingerprints/peptides_molecular_fingerprints_classification https://doi.org/10.5281/zenodo.19388783