Fig. 5.

Performance of 15 Machine Learning Models and Development of HP16118P Biomarker. a Heatmap. From four selected cytokines, a heatmap displays the performance metrics of 15 machine learning models. The QDA model is emphasized for its superior diagnostic capabilities. Key terms: Classif.ce reflects multiclass classification; Kappa denotes Cohen’s coefficient; Accuracy Lower/Upper are confidence intervals; Accuracy Null is the baseline accuracy; Accuracy P Value assesses statistical significance; McNemar P Value compares model performance; Recall, or sensitivity, measures correct positive predictions; and F1 score combines precision and recall values. b Experimental Flowchart: The creation of the multi-epitope biomarker HP16118P advances tuberculosis diagnosis by differentiating ATB from LTBI. The discovery of IL-5 as a specific differentiating cytokine highlights the biomarker’s utility and its potential to enhance immune response, showcasing a significant breakthrough in tuberculosis management and global health impact