AI-guided discovery of fungal-derived salty and saltiness-enhancing peptides for sodium-reduced foods

Natural salty and saltiness-enhancing peptides are promising candidates for sodium reduction in food systems. This study established an integrated AI-guided workflow combining in silico hydrolysis, embedding-based machine learning, electronic tongue validation, cellular evaluation, molecular docking, and molecular dynamics simulation to identify fungal-derived saltiness-modulating peptides. A taste-active peptide dataset was constructed, and Uni-Mol, ESM-3, and MolCLR embeddings were compared using the same FT-Transformer framework. ESM-3 showed the most favorable classification performance and was selected for screening peptides generated from 14 fungal proteomes. Six candidates were synthesized, including the salty peptides AGHQR, AHDQY, and SKPGNR, and the saltiness-enhancing peptides IPVVVDL, DNNEEW, and ANSDL. Electronic tongue analysis identified AHDQY as the strongest salty peptide in deionized water and ANSDL as the most effective saltiness enhancer in a low-salt NaCl system. These peptides exhibited good cytocompatibility and protected HUVECs against high-salt-induced cellular injury to varying degrees, as evidenced by restored nitric oxide (NO) production and reduced reactive oxygen species (ROS) accumulation. Molecular docking and molecular dynamics simulations further supported their stable interactions with the transmembrane channel-like protein 4 (TMC4) and peptide-induced modulation of local receptor flexibility and secondary structure. These findings provide promising fungal-derived peptide candidates and a systematic strategy for developing natural saltiness-modulating ingredients for sodium-reduced foods.