Optimal variable selection for Fourier transform infrared spectroscopic analysis of articular cartilage composition

Articular cartilage (AC) is mainly composed of collagen, proteoglycans, chondrocytes, and water. These constituents are inhomogeneously distributed to provide unique biomechanical properties to the tissue. Characterization of the spatial distribution of these components in AC is important for understanding the function of the tissue and progress of osteoarthritis. Fourier transform infrared (FT-IR) absorption spectra exhibit detailed information about the biochemical composition of AC. However, highly specific FT-IR analysis for collagen and proteoglycans is challenging. In this study, a chemometric approach to predict the biochemical composition of AC from the FT-IR spectra was investigated. Partial least squares (PLS) regression was used to predict the proteoglycan content (n=32) and collagen content (n=28) of bovine cartilage samples from their average FT-IR spectra. The optimal variables for the PLS regression models were selected by using backward interval partial least squares and genetic algorithm. The linear correlation coefficients between the biochemical reference and predicted values of proteoglycan and collagen contents were r=0.923 (p<0.001) and r=0.896 (p<0.001), respectively. The results of the study show that variable selection algorithms can significantly improve the PLS regression models when the biochemical composition of AC is predicted.