First report on ATR-FTIR and machine learning-based sputum profiling for TB detection: A compendium of potential biomarkers.

Tuberculosis (TB) remains a significant global health issue, with millions of new cases reported annually and persistent limitations in conventional diagnostic approaches. This study is the first report that explores the use of Attenuated Total Reflection Fourier-Transform Infrared (ATR-FTIR) spectroscopy for identifying biochemical signatures associated with TB in lyophilized sputum. Spectra were acquired across the full mid-infrared range (400-4000 cm-1) from 100 samples (50 TB-positive, 50 healthy controls) from Pakistan Institute of Medical Sciences, Islamabad, followed by pre-processing and multivariate analysis. Principal Component Analysis (PCA) combined with Linear Discriminant Analysis (LDA) and Support Vector Machine (SVM) classification achieved strong separation between groups. Spectral differences were most pronounced in two regions: 2800-3500 cm-1 and 900-1800 cm-1. Key wavenumbers including 3290 cm-1 (polysaccharides), 2954 cm-1 (mycolic acid), and 1636 cm-1 (β-sheets) have promising potential in establishing disease status. Peaks corresponding to carbohydrates, lipids, and nucleic acids further show the biochemical divergence between the two groups. The observed spectral variations demonstrate a clear potential for sputum-based differentiation of TB using vibrational spectroscopy. These findings highlight the potential for reagent-free, rapid TB screening in clinical and resource-limited settings and contribute to the growing body of research. Future studies on larger data sets can help validate our findings.