Prospection and prediction of highly active antibiofilm peptides using machine learning-based methods

Abstract

Antibiotic resistance is a sign that the golden era of antibiotics is ending. Bacterial biofilm plays a crucial role in the emergence of antibiotic resistance. The biofilms formation on various substrates, from tissues to medical devices, and the remarkable resistance of biofilm-producing bacteria to almost all common antibiotics make bacterial biofilms one of the pivotal challenges in healthcare systems. Finding new therapeutic agents seems inevitable and should be sought proactively. These agents should have particular characteristics to perform well in the biofilm environment. Peptides have been shown to have promising potential as antimicrobial agents. Designing peptides with significant antibiofilm effects is cumbersome and expensive. Developing computational approaches for the prediction of the anti-biofilm effects of peptides seems to be unavoidable. In this study, emphasizing higher than 50% anti-biofilm activity, we applied multiple classification algorithms to select peptide sequences with a considerable anti-biofilm effect for subsequent experimental evaluations. Feature vectors were calculated for each sequence based on the peptide sequences’ primary structure, amino acids’ order, and physicochemical properties. Our computational approach predicted the significant anti-biofilm effect of peptides with accuracy, precision, MCC, and f1-score equal to 99%, 99%, 0.97, and 0.99, respectively, which is comparable with previous methods. This combination of the feature space and high antibiofilm activity was applied in this study for the first time.