Fig. 6
From: microCLIP super learning framework uncovers functional transcriptome-wide miRNA interactions
Assessment of microCLIP prediction efficacy against MIRZA, microMUMMIE, PARma, and Targetscan v7. miRNA-target pairs for each AGO-CLIP in silico approach were obtained from the analysis of 7 PAR-CLIP HEK293 libraries and functional investigation was performed by measuring mRNA responses to miRNA perturbations. Unified sets of a 4 microarray and b 2 RNA-Seq datasets, in which miRNAs were individually transfected into HEK293 cells, were included in the evaluation process. Median fold-change values (log2) of the top predicted targets per tested algorithm were plotted and accordingly compared by applying stepwise cutoffs on total predictions. Performed comparisons additionally incorporate a group comprising mean fold changes of 1000 randomly selected genes (without replacement) by using 100 re-samplings. The gray shaded area represents the minimum-to-maximum log2 fold-change range of the re-samplings per number of top predictions. microCLIP significantly outperforms all the juxtaposed implementations, detecting targets with the strongest median downregulation, from stringent to loose prediction thresholds (two-tailed Wilcoxon signed-rank test)
