Colistin resistance is one of the major threats for global public health, requiring reliable and rapid susceptibility testing methods. The aim of this study was the evaluation of a MALDI-TOF mass spectrometry (MS) peak-based assay to distinguish colistin resistant (colR) from susceptible (colS) Escherichia coli strains. To this end, a classifying algorithm model (CAM) was developed, testing three different algorithms: Genetic Algorithm (GA), Supervised Neural Network (SNN) and Quick Classifier (QC). Among them, the SNN- and GA-based CAMs showed the best performances: recognition capability (RC) of 100% each one, and cross validation (CV) of 97.62% and 100%, respectively. Even if both algorithms shared similar RC and CV values, the SNN-based CAM was the best performing one, correctly identifying 67/71 (94.4%) of the E. coli strains collected: in point of fact, it correctly identified the greatest number of colS strains (42/43; 97.7%), despite its lower ability in identifying the colR strains (15/18; 83.3%). In conclusion, although broth microdilution remains the gold standard method for testing colistin susceptibility, the CAM represents a useful tool to rapidly screen colR and colS strains in clinical practice.
Keywords: Gram-negative bacteria; MALDI-TOF MS; colistin resistance.