Near-infrared spectroscopy and variable selection techniques to discriminate Pseudomonas aeruginosa strains in clinical samples

Abstract Pseudomonas aeruginosa is a leading cause of nosocomial infections, ranking second among the negative Gram pathogens reported to the National Nosocomial Infection Surveillance System. P. aeruginosa may develop resistant during prolonged therapy with all antimicrobial agents. Therefore, isolates that are initially susceptible may become resistant within 3–4 days after initiation of therapy. Testing of repeat isolates may be warranted. There is a need for sensitive and specific tests. We set out to determine whether near-infrared spectroscopy (NIR) combined with variable selection techniques employing successive projection algorithm – linear discriminant analysis (SPA-LDA) or genetic algorithm – (GA-LDA) could discriminate P. aeruginosa strains according to resistant vs. sensitive. The variables selected were then used for discriminating the strains. The influence of various spectral pre-treatments (Savitzky–Golay smoothing, multiplicative scatter correction (MSC), and Savitzky–Golay derivatives) was calculated. In addition, accuracy test results including sensitivity and specificity were determined. Sensitivity in the resistant category was 95% using a SPA-LDA model with 70 wavelengths. Sensitivity and specificity in both categories was 93% using a GA-LDA model with 32 wavelengths. We show that NIR spectroscopy of P . aeruginosa combined with variable selection techniques is a powerful tool for resistant vs. sensitive strains based on the unique spectral “fingerprints” of their biochemical microbial identification, emerging as an alternative for rapid and cost-effective identification of strains.