Comparison of combination and first overtone spectral regions for near-infrared calibration models for glucose and other biomolecules in aqueous solutions.

Partial least squares calibration models are compared for the measurement of glucose, lactate, urea, ascorbate, triacetin, and alanine in aqueous solutions from single-beam spectra collected over the first overtone (6500-5500 cm(-1)) and the combination (5000-4000 cm(-1)) regions of the near-infrared spectrum. Spectra are collected under two sets of conditions with one designed for combination spectra and the other designed for first overtone spectra. As part of the optimization of conditions, an exponential function is presented that accurately characterizes the strong dependency between spectral quality and sample thickness. Sample thickness set for the first overtone and combination spectra are 7.5 and 1.5 mm, respectively. Independent calibration models are established for each solute from both combination and first overtone spectra. Direct comparison reveals superior performance by models generated from combination spectra, particularly for glucose and urea. Standard error of prediction (SEP) values are 1.12 and 0.45 mM for glucose models generated from first overtone and combination spectra, respectively. SEP values for urea are 7.33 and 0.10 mM for first overtone and combination spectra, respectively. Such high SEP values for urea with first overtone spectra correspond to an inability to quantify urea from these spectra because of a lack urea-specific molecular absorption features in this spectral region. Net analyte signal (NAS) is used to quantify the degree of selectivity provided within the first overtone and combination spectral regions. The superior selectivity of combination spectra is confirmed by comparing the length of the NAS vectors for each matrix component.