Combining statistical, machine learning and experimental approaches for screening of novel antimicrobial peptides from complex hydrolysates
Laurent Bazinet
Université Laval, Québec, Canada
Producing bioactive peptides through enzymatic hydrolysis is one of the most promising strategies for valorizing food by-products. Besides the optimization of critical process parameters, the characterization of hydrolyzed proteins is one of the key steps in the development of peptide-based bioactive ingredients. The complexity of the raw hydrolysates produced, which contain a diverse mixture of peptides with different levels of abundance, presents a significant challenge for the identification of those peptides that contribute to the observed biological activity. This study presents for the first time an integration of conventional statistical and machine learning tools to discover new antimicrobial peptides from complex hydrolysates, based on experimental data of the antimicrobial activity of raw hydrolysates and their peptide population. Partial-Least Square Discriminant Analysis (PLS-DA), Pearson correlation (P-corr), Linear Regression (LR), and Random Forest (RF) were used to explain the relationship between peptide population abundance and antimicrobial activities (antibacterial, anti-mold, and anti-yeast) of raw hydrolysates. Peptides having greater importance in explaining the antimicrobial activities of hydrolysates were selected and their in vitro antimicrobial activity was further assessed by chemical synthesis. As a result, new antimicrobial peptide sequences were identified. This innovative approach can accelerate the discovery of new antimicrobial peptides from complex hydrolysates, which could be useful for further separation and application of peptides in food biopreservation.