Figure 4.

Prediction of TB treatment outcome in dcRT-MLPA data from peripheral blood. (a) ROC curves showing the predictive power of the gene signatures identified in the balanced pooled cohort (South Africa and Indonesia) to classify TB patients at diagnosis (left panel), two weeks (middle panel) or two months (right panel) after initiation of TB treatment into patients who had a good TB treatment outcome and patients who had a poor TB treatment outcome, using the RFE-RF model and LOOCV. The dataset was balanced by down-sampling to encompass the same number of individuals with poor and good TB treatment outcome (diagnosis, n = 34; week two, n = 33; month two, n = 34). (b) Gene expression kinetics of the single genes encompassing the diagnosis or week two gene signatures predicting TB treatment outcome in the pooled cohort. Box plots depict GAPDH-normalized, log₂-transformed median gene expression values and the IQR, while the whiskers represent the data within the Q₁-1·5xIQR and Q₃+1·5xIQR interval. (c). ROC curves showing the predictive power of the gene signatures identified in the balanced pooled cohort (South Africa and Indonesia) to classify TB patients from an external validation cohort (India) at diagnosis (left panel) or two months (right panel) after initiation of TB treatment into patients who had a good TB treatment outcome and patients who had a poor TB treatment outcome. The dataset was balanced by down-sampling to encompass the same number of individuals with poor and good TB treatment outcome (diagnosis, n = 22; month two, n = 22). Abbreviations: AUC, area under the curve; CI, confidence interval.