ROC curves of the prediction performances. (A) 100 Support Vector Machines were trained with the datasets ecoB and ecoG, respectively, and were then queried using the datasets from P. aeruginosa (union of the datasets paeL and paeJ). The number of machines predicting essentiality was summed up (voting score). Results from varying thresholds of the voting score were compared to the experimental results of paeL and paeJ yielding the ROC curves (area under the curve: 0.80 and 0.79, respectively). (B) Similar to (A) only that the machines were trained with the datasets of P. aeruginosa and queried with the datasets of E. coli resulting in ROC curves with AUC = 0.81 and 0.75 for the datasets ecoB and ecoG, respectively.