Fig. 2

Comparison of prediction performance of different predictors in eleven newly-designed datasets under multi-fold cross validation. X-axis shows the eleven newly-designed datasets, while the y-axis shows (a) accuracy, (b) sensitivity, and (c) F1 score, respectively. Error bars are standard deviation from multi-fold cross validation. In the D4 dataset, the performance of mirTarDANN was compared with four individual predictors and ComiR. In each of the D3 series datasets, mirTarDANN was compared to ComiR, and three out of four individual predictors. In each of the D2 series datasets, mirTarDANN was compared to ComiR and two out of four individual predictors. When calculating the accuracy of individual predictors, their default cutoff values were used. For ComiR, a false discovery rate of 5% was recommended by the developer to determine the cutoff. Therefore, based on the calculations of 50 randomly selected miRNAs and their targets in the datasets, 0.82 was used as the cutoff of ComiR