Effective identification of clinical bacterial pathogens by fourier transform near-infrared spectroscopy

The Escherichia coli (E. coli, ATCC 25922), Staphylococcus aureus (S. aureus, ATCC 29213), and Salmonella (SE, ATCC 14028) are three common bacterial pathogens of BSIs (Bloodstream infection). Accurately identifying these three bacterial pathogens will greatly help doctors to reduce the number of days to cure the patients. In this study, the identification models for bloodstream infection are studied. Firstly, Fourier transform near-infrared spectroscopy (FT-NIR) and multivariate calibration method are applied to detect and discriminate these bacterial suspension samples. Four preprocessing methods (multiplicative scatter correction (MSC), standard normal variate correction, Savitzky-Golay first-derivative, and Savitzky-Golay second-derivative are adopted to modify the raw spectral data. Then, three discriminant models are built to distinguish unknown bacterial samples, they are the model based on the principal component analysis and mahalanobis distance discriminant (PCA-MDD), the model based on the partial least squares-discriminant analysis (PLS-DA), and the back propagation neural network model. Finally, the effectiveness of the pre-processing methods and models are discussed and the comparison results are given. The results indicate that MSC is the most useful pre-processing method for the bloodstream infection spectra. By using MSC, the PLS-DA model and the BP model obtain higher accuracy than PCA-MDD. Moreover, although both the prediction accuracy of the PLS-DA model and the BP model are 100%, the difference between the predicted values and the real value are the minimal in the BP model. Thus BP model has better performance. Hence, FT-NIR spectroscopy combined with chemometrics techniques can be a useful, rapid, and nondestructive tool to discriminate clinical bacterial pathogens.