Machine learning–driven prediction of antimicrobial resistance in Staphylococcus aureus using genomic and allelic signatures

Abstract This novel study leveraged advanced machine learning techniques to elucidate the molecular mechanisms of antimicrobial resistance (AMR) in 300 Staphylococcus aureus isolates across six critical antibiotics. Employing a diverse array of deep learning and ensemble models, we conducted an in-depth analysis of genetic markers and allelic variations to characterize resistance determinants. Our investigation revealed that the XGBoost ensemble model demonstrated the most exceptional performance, achieving a remarkable 95% test accuracy, 100% training accuracy, and an unprecedented ROC AUC of 0.9855. Comparative analysis of multiple machine learning approaches, including Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN), Multi-Layer Perceptron (MLP), Decision Tree, and Stochastic Gradient Descent (SGD) models, provided detailed insights into resistance prediction. The SHAP (SHapley Additive exPlanations) analysis unveiled critical genetic markers, with “cat_allele_Cluster_1015_Allele_8” emerging as the most influential feature driving resistance predictions. Notably, the models exhibited varying performance across different antibiotics, with consistently high accuracy and F1-scores for ciprofloxacin, clindamycin, gentamicin, and sulfamethoxazole/trimethoprim. Our findings not only demonstrate the potential of advanced machine learning techniques in predicting antimicrobial resistance but also provide crucial insights into the molecular mechanisms underlying S. aureus drug resistance. By identifying key genetic determinants and their relative importance, this study offers a sophisticated approach to understanding resistance patterns, potentially guiding future diagnostic strategies, targeted therapies, and antimicrobial stewardship practices in clinical settings.