A machine learning-assisted analysis of essential oil constituents targeting salmonella typhimurium

Some essential oils (EOs) have exhibited high efficacy against Salmonella Typhimurium, suggesting that specific molecular structures within their constituents could inform the development of alternative antimicrobial agents, potentially addressing the rise of antibiotic resistance in this bacterium. Through feature permutation analysis on a dataset of 171 EO samples encompassing 682 molecular substructures, we identified ten key predictors of antibacterial activity against S. Typhimurium. These predictors were used to train a logistic regression-based machine learning model. Hydroxyl-substituted benzene rings characteristic of phenolic compounds - such as carvacrol, thymol and eugenol - emerged as strong predictors of antibacterial activity. In contrast, non-aromatic bicyclic structures present in monoterpenoids like alpha-pinene, beta-pinene and delta-3-carene were associated with a lack of efficacy against Salmonella. The molecular features identified align with existing research on the antimicrobial properties of phenolic compounds, thereby validating the use of machine learning approaches in guiding the discovery of naturally occurring antimicrobial agents.