AI, docking, and molecular dynamics to track the binding of structural peptides to different keratin models

The present work shows a computational approach to assess the interactions of different nature-inspired peptides with hair keratin models. An updated keratin model was validated, and comparisons with previous models were traced, thereby highlighting the time-evolution of both the model and the in silico methods. Importantly, the computational methodology presented here allows for the study of peptide-protein interactions between unknown or unsolved structures. For that, artificial intelligence, molecular docking, and molecular dynamics simulations were used to predict the peptides' structural conformation, stability, binding, and interaction patterns with keratin. The keratin models were designed to include a conserved form and two hyperoxidized forms, where a few thiols were replaced by sulphonic acid in cysteine residues. Most peptides displayed strong affinity and stability over time, established by distinct interactions with the three keratin models, but peptides P34 (resilin-like) and P36 (abductin-like) stood out due to their amino acid variety, the different types of interactions they performed and the sustained binding proximity within the keratin helices. AlphaFold also demonstrated high reliability and accuracy when constructing the 'virgin' keratin model and the peptide structures. This study furthers many others on peptide-protein interactions whilst paving another setting stone in the use of computational methods on previously under-explored biotechnological fields such as cosmetics.