Fig. 7

microCLIP performance compared to MIRZA, microMUMMIE, PARma, and Targetscan v7 was examined in seven public gene expression profiling datasets following miRNA transfection or knockdown in HEK293 and HeLa cell lines. miRNA-target interactions for AGO-CLIP in silico approaches were obtained from the analysis of PAR-CLIP HEK293 and HeLa libraries reported in the studies of Kishore et al. and Whisnant et al. Response of targeted mRNAs to miRNA perturbation experiments was evaluated independently per tested cell type, experimental technique and condition (a–g). Cumulative distributions of mRNA fold changes for targets comprising at least one predicted MRE in the CDS or 3′ UTR regions were compared to those that lacked any site of the considered miRNAs (one-sided Kolmogorov–Smirnov test). Functional efficacy was assessed for equal numbers of top predictions per implementation. Implementations that did not support targets with a fold-change in the examined miRNA perturbation experiments were not included in the relevant cumulative plots. a–f Identified targets by microCLIP revealed greater site effectiveness than the rest AGO-CLIP-guided implementations. g microCLIP performed similarly as PARma and better than the rest of implementations. Targetscan v7 identifies responsive targets, operating on par with in silico approaches based on CLIP data such as PARma, while in c, d and g it displays analogous efficacy as microCLIP. The number of transcripts included in each comparison is denoted in the parentheses. Log₂-transformed expression fold-change values of all perturbation experiments used in the comparisons are provided in Supplementary Data 2