Investigation of sputum volatiles to classify active tuberculosis: A pilot study

Tuberculosis (TB) remains a major global health challenge due in part to limitations in rapid and affordable diagnostics. Current diagnostic methods are time-intensive and often inaccessible in resource-limited settings, emphasizing the urgent need for rapid, low-cost screening approaches. One promising strategy involves the analysis of volatile molecules associated with TB-infection. In this study (n = 100) we identify 14 sputum-derived volatiles and utilize them to construct a machine learning model that classifies samples by TB status with a sensitivity of 90% and a specificity of 86% across cross-validation folds. The resulting profile provides a foundation for further biomarker validation studies with an expanded sample size and the development of non-invasive breath diagnostics.