Data visualization of Salmonella Typhimurium contamination in packaged fresh alfalfa sprouts using a Kohonen network.

Class visualization of multi-dimensional data from analysis of volatile metabolic compounds monitored using an electronic nose based on metal oxide sensor array was attained using a Kohonen network. An array of 12 metal oxide based chemical sensors was used to monitor changes in the volatile compositions from the headspace of packaged fresh sprouts with and without Salmonella Typhimurium contamination. Kohonen׳s self-organizing map (SOM) was then created for learning different patterns of volatile metabolites. The Kohonen network comprising 225 nodes arranged into a two-dimensional hexagonal map was used to locate the samples on the map to facilitate sample classification. Graphical maps including the unified matrix, component planes, and hit histograms were described to characterize the relation between samples. The clustering of samples with different levels of S. Typhimurium contamination could be visually distinguishable on the SOM. The Kohonen network proved to be advantageous in visualization of multi-dimensional nonlinear data and provided a clearer separation of different sample groups than a conventional linear principal component analysis (PCA) approach. The sensor array integrated with the Kohonen network could be used as a rapid and nondestructive method to distinguish samples with different levels of S. Typhimurium contamination. Although the analyses were performed on samples with natural background microbiota of about 7 Log(CFU/g), this microbiota did not affect the S. Typhimurium detection. The proposed method has potential to rapidly detect a target foodborne pathogen in real-life food samples instantaneously without subsequently culturing stages.