Identification of bacterial pathogens using quadrupole mass spectrometer data and radial basis function neural networks

A quadrupole mass spectrometer has been employed to analyse the headspace above bacterial cultures. This, along with a pattern recognition algorithm, constitutes an electronic nose system. Here we present the results of a study on the headspace of pathogens, specifically Escherichia coli K12 and Staphylococcus aureus, the purpose being to identify the growth phase and strain of different pathogens. The data collected from the mass spectrometry were used to train a radial basis function (RBF) neural network. This type of network was employed because it requires smaller training sets and is suitable for what is, in effect, 505 mass ´sensors´. Principal components analysis shows that there is sufficient information in the volatiles to discriminate between the different growth phases of E. coli, but less so for two strains of S. aureus, i.e. MRSA and NCTC. Excellent results are obtained using these RBF neural networks as approximators of discriminant functions. Furthermore, it is demonstrated that this method can deal with classification problems that involve nonlinearity in the data. It is concluded that the reported methodology shows promise as a useful pathogen identification technique, and in particular discrimination between the virulent MRSA and the innocuous NCTC strain.