*Tentative band assignment references.[18,20]
3.2 Principal Component Analysis
MVA statistical routines were applied to the raw spectroscopic data of
the microorganisms under study. The use of chemometrics in spectroscopy
has been well documented in the MIR and Raman spectroscopies
[20,22-23]. The MVA routine allows determining the variables
responsible for separating and classifying the classes. For this work,
the classification of the microorganisms was evaluated by performing
pre-processing algorithms to reduce common scattering and background
effects that are inevitable during spectral acquisition. It is common to
see spectral variation effects due to sample heterogeneity, scattering
attributed to particle size, solvent vapor pressure, and analyte
solubility. These effects are corrected with the pre-processing
algorithms, which allow correction in the data for accurate
classification. A comparison between the raw unprocessed spectral data
and pre-processed is shown in Appendix A, Figure 1A.
Principal Component Analysis (PCA) models were generated for the
microorganisms considered for this discussion as neat and mixtures of
these analytes representing a complex environment. The models were built
on vibrational information using the 788 - 1884 cm-1spectral region. First, the PCA model for the neat Ec , Bt ,
and Se bacteria is shown in Figure 5. This model was generated
using the Standard Normal Variate (SNV) processing algorithm. It is
represented by the scores of principal components 1 (PC-1) vs. PC-2 for
the bacteria deposited on SS. PC-1 and PC-2 scores represented the
largest percentage of variation for the acquired spectra. The PC-1
accounts for 52% of the total spectral variance and 19% for the PC-2.