Discrimination of species within the Enterobacter cloacae complex using MALDI-TOF Mass Spectrometry and Fourier-Transform Infrared Spectroscopy coupled with Machine Learning tools
The Enterobacter cloacae complex (ECC) encompasses heterogeneous clusters of species that have been associated with nosocomial outbreaks. These species may host different acquired antimicrobial resistance and virulence mechanisms and their identification are challenging. This study aims to develop predictive models based on MALDI-TOF MS spectral profiles and machine learning for species-level identification. A total of 198 ECC and 116 K. aerogenes clinical isolates from the University Hospital Ramon y Cajal (Spain) and the University Hospital Basel (Switzerland) were included. The capability of the proposed method to differentiate the most common ECC species (E. asburiae, E. kobei, E. hormaechei, E. roggenkampii, E. ludwigii, E. bugandensis) and K. aerogenes was demonstrated by applying unsupervised hierarchical clustering with PCA pre-processing. We observed a distinctive clustering of E. hormaechei and K. aerogenes and a clear trend for the rest of the ECC species to be differentiated over the development dataset. Thus, we developed supervised, non-linear predictive models (Support Vector Machine with Radial Basis Function and Random Forest). The external validation of these models with protein spectra from the two participating hospitals yielded 100% correct species-level assignment for E. asburiae, E. kobei, and E. roggenkampii and between 91.2% and 98.0% for the remaining ECC species. Similar results were obtained with the MSI database developed recently (https://msi.happy-dev.fr/) except in the case of E. hormaechei, which was more accurately identified by Random Forest. In short, MALDI-TOF MS combined with machine learning demonstrated to be a rapid and accurate method for the differentiation of ECC species.