Fig 1.
Workflow showing data pre-processing steps (A) and regression modeling (B).
PID: participant ID; AUC: area under the curve; LR: logistic regression; CB: candidate biomarkers; CV: cross validation; LRT: likelihood ratio test.
Fig 2.
Candidate biomarkers of recent M.tb infection identified by the adaptive and innate EN models.
Boxplots comparing the raw values of recent (red) and persistent (blue) QFT+ individuals for the three candidate biomarkers with non-zero coefficients in the adaptive EN model (A) and the 10 biomarkers with non-zero coefficients in the innate EN model (B). Wilcoxon tests were used to compare the two groups and the resulting p-values are shown.
Table 1.
Model estimates and the average performance metrics after internal validation of the final LR model (i) and the LR model built to E6C10-specific total Th 1 cells expressing HLA-DR only (ii).
Fig 3.
(A) Results from the simple classification tree built to the entire integrated dataset. Boxplots comparing vast scaled values of recent (red) and persistent (blue) QFT+ individuals were plotted for the two most stratifying features identified by the decision tree. The split values are superimposed onto the plots at (B) -0.098 for proportions of E6C10-specific Th1 cells expressing HLA-DR, and (C) -0.12 for frequencies of Esp-specific IL2+CD107-CD154-IFN-γ-TNF+ CD4+ T cells. (D) Variable importance plot of the final RF model showing the top 10 variables that resulted in the largest average decrease in the Gini Index.