Serotyping Salmonella Enteritidis and Typhimurium Using Whole Cell Matrix Assisted Laser Desorption Ionization – Time of Flight Mass Spectrometry (MALDI-TOF MS) through Multivariate Analysis and Artificial Intelligence

Abstract

AbstractSalmonella is one of the most frequent food-borne zoonoses, while Salmonella Typhimurium and Enteritidis are the major serovars of concern in public health. 113 Salmonella strains including 38 S. Enteritidis (SE), 38 S. Typhimurium (ST) and 37 strains of 32 other Salmonella serovars (SG) were tested in quadruplicate by whole-cell MALDI-TOF MS. Ions were studied and aligned from the raw data of mzXML files using Mass-Up (http://www.sing-group.org/mass-up), resulting in 1,741 aligned peaks. Datasets of ions (presence/absence) selected using a home-developed criteria on their specificity and detectability were subjected to multivariate analyses and artificial intelligence tools. Principle Component Analysis based on 88 selected ions separated SE, ST and SG without overlap on the first three principle components. The network and forest based deep learning tools were more sophisticated than the decision tree-based models. Neural Network carried out consistently well in training model, but no advantage was gained over the other models in validation results. HP (high performance) Neural, Support Vector Machine, HP Forest and Gradient Boasting were able to identify SE, ST and SG up to 100% correctly in both training and validation when 88 selected ions were used in analysis. Among them, HP Neural seemed to perform slightly better and relatively stable. Selection of serovar specific ions helps develop serotyping by increasing signal to noise. MALDI-TOF MS used with appropriate data processing strategies and classification tools could be applied to quickly alert when Salmonella serotypes of concern are suspected among routinely processed samples.