Fig. 6

Cross-platform predictions. A PCA plot and (B) boxplot on singscores for two selected signatures: TIS and PIP PD-1, for all 161 samples (126 NanoString training samples, 35 WTS testing samples). The singscores were calculated by the “No stable gene” method in the NanoString platform and the “part” method in the WTS platform. The dots are colored by the platform (NanoString or WTS). The p-values were measured by Mann Whitney Wilcoxon test. C ROC curves of predicting NanoString and WTS samples using logistic regression models. Due to using the same model, the “all” and “part” have identical ROC curves when predicting samples from NanoString platform. The colors differentiate the pairs of comparisons. D, E MCC, sensitivity and specificity values of binary clustering based on the predictive probability from different logistic regression models. D using default threshold (probability of response ≥ 0.5 as responder); E using optimal threshold (maximizing Youden index in ROC curve from training samples) in each model. NA in sensitivity was due to no sample being classified as responder