Fig. 3

ROC curves of the Yeast dataset comparing our pipeline with a FDR corrected t-test, and comparing filtering features with missing values with imputing missing values. a shows the performance of limma against FDR corrected t-test on the yeast dataset where all features with missing data have been filtered out. b compared our pipeline using multiple imputation and limma against t-test on a single imputation. Both, LFC, and P-val indicates which parameters were varied (p-value and LFC at the same time, LFC only, and p-value only, respectively) when creating the ROC curves. LFC was changed from zero to two and p-value from 0.05 to 0.001, simultaneously or separately. When only LFC was changed the p-value was fixed to 0.05, and when the p-value was changed LFC was fixed to zero. Lower LFC cut-off was set to –infinity as there was no TP with a decreasing fold change. “Limma” indicates that our pipeline utilizing limma was used with 100 imputations while “t-test” indicates that a FDR corrected t-test was used on a single imputed dataset. Y-axis show TPR and x-axis show FPR expressed as a percentage (TPR*100, FPR*100, respectively)