Lindsey, B.; Bowden, K.; Shaul, Y.; Petricoin, E.; Caswell, S. V.; Alhammad, R.; Elayadi, A. N.; Martin, J.

Introduction Physical fatigue is a key determinant of operational readiness in the physically demanding occupations of tactical athletes. Specific hormonal, immune, and enzymatic biomarkers have been proposed for fatigue assessment, but their reliability can be affected by external factors. This study aimed to explore the predictive accuracy of targeted stress-related small molecules to proteins identified via untargeted salivary proteomics in classifying physical fatigue. Methods Ten recreationally active adults (6M, 4F) completed a fatiguing protocol designed to simulate common operational movements and intensities. Saliva samples were collected pre- and post-protocol and analyzed for targeted biomarkers using commercial immunoassays and for untargeted proteins via liquid chromatography-mass spectrometry. Machine learning models were trained to classify pre- vs post-exercise state using these biomarkers, with performance assessed through sensitivity, specificity, and area under the receiver operating characteristic curve (AUC). Results Targeted small molecules achieved an overall model accuracy of 86%, with immunoglobulin A and uric acid demonstrating the highest predictive power. However, a proteomic panel of four proteins (ATP1B1, STOML2, PGLYRP2, FH) exhibited superior performance, with 95% classification accuracy and improved sensitivity. Pathway analysis revealed that these proteins were involved in mitochondrial function, immune regulation, and metabolic adaptation, suggesting their role in fatigue-associated physiological changes. Conclusions Salivary proteomics identified biomarkers with greater sensitivity and specificity for detecting physical fatigue than targeted stress-related molecules within this sample. These findings support the potential for non-invasive proteomic monitoring of fatigue in operational and athletic settings. Future studies should validate these findings in larger, more diverse populations and assess their applicability to chronic fatigue monitoring.