RESUMO
INTRODUCTION: To date, limited data exist on demonstrating the usefulness of machine learning (ML) algorithms applied to MALDI-TOF in determining colistin resistance among Klebsiella pneumoniae. We aimed to detect colistin resistance in K. pneumoniae using MATLAB on MALDI-TOF database. MATERIALS AND METHODS: A total of 260 K. pneumoniae isolates were collected. Three ML models, namely, linear discriminant analysis (LDA), support vector machine, and Ensemble were used as ML algorithms and applied to training data set. RESULTS: The accuracies for the training phase with 200 isolates were found to be 99.3%, 93.1%, and 88.3% for LDA, support vector machine, and Ensemble models, respectively. Accuracy, sensitivity, specificity, and precision values for LDA in the application test set with 60 K. pneumoniae isolates were 81.6%, 66.7%, 91.7%, and 84.2%, respectively. CONCLUSION: This study provides a rapid and accurate MALDI-TOF MS screening assay for clinical practice in identifying colistin resistance in K. pneumoniae.