Curve fitting in Fourier transform near infrared spectroscopy used for the analysis of bacterial cells

Infrared spectroscopy is a prominent molecular technique for bacterial analysis. Within its context, near infrared spectroscopy in particular brings benefits over other vibrational approaches; these advantages include, for example, lower sensitivity to water, high penetration depth and low cost. However, near infrared spectroscopy is not popular within microbiology, because the spectra of organic samples are difficult to interpret. We propose a comparison of spectral curve-fitting methods, namely, techniques that facilitate the interpretation of most peaks, simplify the spectra and improve the prediction of bacterial species from the relevant near infrared spectra. The performances of three common curve-fitting algorithms and the technique based on the differential evolution were compared via a synthesized experimental spectrum. Utilizing the obtained results, the spectra of three different bacterial species were curve fit by optimized algorithm. The proposed algorithm decomposed the spectra to specific absorption peaks, whose parameters were estimated via the differential evolution approach initialized through Levenberg-Marquardt optimization; subsequently, the spectra were classified with conventional procedures and using the parameters of the revealed peaks. On a limited data set, the correct classification rate computed by partial least squares discriminant analysis was 95%. When we employed the peak parameters for the classification, the rate corresponded to 91.7%. According to the Gaussian formula, the parameters comprise the spectral peak position, amplitude and width. The most important peaks for bacterial discrimination were identified by analysis of variance and interpreted as N-H stretching bonds in proteins, cis bonds and CH2 absorption in fatty acids. We examined some aspects of the behaviour of standard curve-fitting algorithms and proposed differential evolution to optimize the fitting process. Based on the correct use of these algorithms, the near infrared spectra of bacteria can be interpreted and the full potential of near infrared spectroscopy in microbiology exploited.