Estimation and prediction of the air–water interfacial tension in conventional and peptide surface-active agents by random Forest regression

Abstract

The surface tension at the critical micelle concentration (STCMC) is an important descriptor for surfactants in applications including cosmetics, pharmaceuticals and food. A predictive STCMC Random Forest model was trained with 691 conventional surfactants and 9 amino acids. The model was evaluated by fivefold cross-validation, and the prediction power was tested for peptides by direct comparison with the experimental STCMC values found in the literature or measured in this work. Predictions were also conducted for short peptide permutations. The estimated STCMC approached the experimental values as the peptide length decreased, suggesting a strong influence of secondary structures for longer sequences where the developed algorithm fails to make robust predictions. Concerning the short peptide permutations, the model estimated lower STCMC values as the carbon number increased within the hydrophobic portions. This highlights the importance of the hydrophobic amino acids leucine, isoleucine, and phenylalanine in peptides with attractive surfactant properties.