Large-scale AI-guided discovery of antifreeze peptides reveals functional tandem-repeat architectures

AFPFusionLM predicts antifreeze proteins with AUROC 0.937 and AUPRC 0.957 on an independent reference set, enabling AI-guided antifreeze peptide discovery beyond known scaffolds.

• Why it matters: Antifreeze peptide discovery is limited by narrow candidate sequence diversity and dependence on existing protein scaffolds, constraining biocompatible cryopreservation material development.
• What they did: The framework combines multimodal antifreeze protein prediction, antifreeze-domain-adapted generative sequence expansion, and local sequence-pattern analysis to screen UniProt, glacier metagenomes, and generated sequences, yielding 87,305 candidate protein records.
• The result: Recurring periodic tandem-repeat architectures were found across multiple candidate clusters, and experimental testing identified ice recrystallization inhibition in multiple repeat-associated short-peptide families, with cryoprotective support in a representative formulation under reduced-DMSO conditions.

AI summary of the abstract · may contain errors · not medical advice · read the paper.