Predicting antimicrobial properties of lignin derivatives through combined data driven and experimental approach

Abstract

Meta-analysis, experimental and data-driven quantitative structure–activity relationship (QSAR) models were developed to predict the antimicrobial properties of lignin derivatives. Five machine learning algorithms were applied to develop QSAR models based on the ChEMBL, a public non-lignin specific database. QSAR models were refined using ordinary-least-square regressions with a meta-analysis dataset extracted from literature and an experimental dataset. The minimum inhibition concentration (MIC) values of compounds in the meta-analysis dataset correlate to classification-based descriptors and the number of aliphatic carboxylic acid groups (R2 = 0.759). Comparatively, QSARs derived from the experimental datasets suggest that the number of aromatic hydroxyl groups were better predictors of Bacterial Load Difference (BLD, R2 = 0.831) for Bacillus subtilis, while the number of alkyl aryl groups were the strongest correlation in predicting the BLD (R2 = 0.682) of Escherichia coli. This study provides insights into the type of descriptors that correlate to antimicrobial activity and guides the valorization of lignin into sustainable antimicrobials for potential applications in food preservation, fermentation, and other industrial sectors.