Deep Learning Enables Identification of Antimicrobial Peptides Through Mechanochromic Fingerprints

Antimicrobial peptides (AMPs) are promising antibiotic alternatives, but their diverse modes of action make functional classification slow and labor-intensive. Here we introduce a rapid and scalable strategy for AMP identification that integrates low-cost, self-assembled polydiacetylene (PDA) sensors with hyperspectral imaging and deep learning. AMP-PDA interactions generate mechanochromic spectral fingerprints that capture subtle differences in affinity, conformation, and penetration depth in membranes. Convolutional neural networks (CNNs) trained on full spectral datasets accurately distinguished seven AMPs at two concentrations with 96.79% accuracy, whereas conventional two-wavelength colorimetric response failed entirely. The rich chemical information embedded across the entire visible wavelength spectral region reveals that mechanochromic polymers encode far more detail than previously recognized. These findings establish PDA mechanochromism, when paired with high-throughput spectral imaging and deep learning, as a powerful and accessible platform for rapid AMP screening and, more broadly, as a foundation for scalable, information-dense biosensing technologies.