Attenuated Total Reflectance-Fourier Transform Infrared (ATR-FTIR) Spectroscopy Discriminates the Elderly with a Low...

In the aging process, the presence of interleukin (IL)-17-producing CD4+CD28-NKG2D+T cells (called pathogenic CD4+ T cells) is strongly associated with inflammation and the development of various diseases. Thus, their presence needs to be monitored. The emergence of attenuated total reflectance-Fourier transform infrared (ATR-FTIR) spectroscopy empowered with machine learning is a breakthrough in the field of medical diagnostics. This study aimed to discriminate between the elderly with a low percentage (LP; ≤3%) and a high percentage (HP; ≥6%) of pathogenic CD4+CD28-NKG2D+IL17+ T cells by utilizing ATR-FTIR coupled with machine learning algorithms. ATR spectra of serum, exosome, and HDL from both groups were explored in this study. Only exosome spectra in the 1700-1500 cm^-1 region exhibited possible discrimination for the LP and HP groups based on principal component analysis (PCA). Furthermore, partial least square-discriminant analysis (PLS-DA) could differentiate both groups using the 1700-1500 cm^-1 region of exosome ATR spectra with 64% accuracy, 69% sensitivity, and 61% specificity. To obtain better classification performance, several spectral models were then established using advanced machine learning algorithms, including J48 decision tree, support vector machine (SVM), random forest (RF), and neural network (NN). Herein, NN was considered to be the best model with an accuracy of 100%, sensitivity of 100%, and specificity of 100% using serum spectra in the region of 1800-900 cm^-1. Exosome spectra in the 1700-1500 and combined 3000-2800 and 1800-900 cm^-1 regions using the NN algorithm gave the same accuracy performance of 95% with a variation in sensitivity and specificity. HDL spectra with the NN algorithm also showed excellent test performance in the 1800-900 cm^-1 region with 97% accuracy, 100% sensitivity, and 95% specificity. This study demonstrates that ATR-FTIR coupled with machine learning algorithms can be used to study immunosenescence. Furthermore, this approach can possibly be applied to monitor the presence of pathogenic CD4+ T cells in the elderly. Due to the limited number of samples used in this study, it is necessary to conduct a large-scale study to obtain more robust classification models and to assess the true clinical diagnostic performance.