Combining Results from Distinct MicroRNA Target Prediction Tools Enhances the Performance of Analyses

Abstract

Target prediction is generally the first step toward recognition of bona fide microRNA (miRNA)-target interactions in living cells. Several target prediction tools are now available, which use distinct criteria and stringency to provide the best set of candidate targets for a single miRNA or a subset of miRNAs. However, there are many false-negative predictions, and consensus about the optimum strategy to select and use the output information provided by the target prediction tools is lacking. We compared the performance of four tools cited in literature—TargetScan (TS), miRanda-mirSVR (MR), Pita, and RNA22 (R22), and we determined the most effective approach for analyzing target prediction data (individual, union, or intersection). For this purpose, we calculated the sensitivity, specificity, precision, and correlation of these approaches using 10 miRNAs (miR-1-3p, miR-17-5p, miR-21-5p, miR-24-3p, miR-29a-3p, miR-34a-5p, miR-124-3p, miR-125b-5p, miR-145-5p, and miR-155-5p) and 1,400 genes (700 validated and 700 non-validated) as targets of these miRNAs. The four tools provided a subset of high-quality predictions and returned few false-positive predictions; however, they could not identify several known true targets. We demonstrate that union of TS/MR and TS/MR/R22 enhanced the quality of in silico prediction analysis of miRNA targets. We conclude that the union rather than the intersection of the aforementioned tools is the best strategy for maximizing performance while minimizing the loss of time and resources in subsequent in vivo and in vitro experiments for functional validation of miRNA-target interactions.

Improving Bioinformatics Prediction of microRNA Targets by Ranks Aggregation

“…This is in line with multiple studies that combine data to obtain the most relevant interactions (Shirdel et al, 2011;Marbach et al, 2012;Friedman et al, 2014;Andres-Leon et al, 2015;Sedaghat et al, 2018). A recent study in particular shows that the union of the predictions of three tools among four (TargetScan, miRanda-mirSVR, RNA22) increases the performance of the analyses (Oliveira et al, 2017). However, our work goes further since prediction lists were aggregated and reranked in a unique list.…”

“…Interestingly, it has been demonstrated that targets resulting from the intersection of two lists of predictions are not more likely to be present in the intersection of two other lists (Ritchie et al, 2009). Therefore, intersecting results does not increase the probability of retaining true positives and it may lead to decreased sensitivity because of possibly omitting valid interactions (Sethupathy et al, 2006;Oliveira et al, 2017). In order to circumvent these limitations, we computed a new score based on the aggregation of the interaction ranks taken from other well-known prediction algorithms.…”

Genetic variability of serotonin pathway associated with schizophrenia onset, progression, and treatment

“…In this chapter, miRNA genes are discussed more precisely, as they are better studied than other ncRNA genes involved in SZ. Moreover, miRNAs are easier to associate with metabolic pathways, as miRNA‐target binding patterns are relatively well known and many miRNAs already have experimentally proven targets (Lee, Kim, Muth, & Witwer, ; Oliveira et al, ). Thus miRNAs present good candidates for investigation of regulation of 5‐HT pathway in SZ.…”

Review of Computational Prediction of Competing Endogenous RNA

“…Moreover studies reveal that combinations of target prediction tools have different level of performance. In a study by [75] the union of Target Scan and MiRanda-mirSVR showed good performance in terms of specificity and precision, and the union of TargetScan, MiRand-mirSVR and RNA22 offered remarkable sensitivity. Hence there is a scope for experimenting with different combination of tools in enhancing ceRNA prediction accuracy.…”