Abstract
The quantitative analysis of fatty acids (FAs) and their acyl group compositions in triacylglycerols (TAGs) has become one of the main areas of interest for understanding the metabolism and function of fats in the body. Although Raman spectroscopy and chemometric-based analytical methods have previously been applied for directly and nondestructively analyzing fats, fat samples are difficult to quantitatively analyze because an appropriate analytical model must be constructed based on a known calibration data set before applying the model to unknown samples. Therefore, we developed a technique to construct calibration models for fatty acyl groups using simulated TAG spectra generated from fatty acid methyl esters (FAMEs) spectra. Because of the vast diversity of TAGs and high prices of commercial pure reagents, the preparation of accurate concentrations of TAGs for training models would be very difficult and costly. Classical and nonnegative least-squares regressions (CLSR and NNLSR, respectively), which do not require calibration modeling, were compared with analyses using partial least-squares regression (PLSR). A comparative analysis revealed that the combination of PLSR modeling with simulated calibration data sets produced the most accurate predictions. The PLSR models were evaluated using edible oils and, compared to the results obtained using gas chromatography, the models reasonably approximated the fatty acyl group compositions in the fat samples. Then, the models were applied to estimate fatty acyl group compositions in live, single adipocytes. Although the models' accuracies were limited, they nondestructively estimated the fatty acyl group compositions of LDs in live cells.