A Clustering-Based Approach for Filtering False Positive MicroRNA Candidates

Abstract

Our study first validated the phenomenon of microRNA (miRNA) clustering in the human genome using computational methods, and then showed that miRNA clustering can be used to improve the computational predictions of human miRNAs. We demonstrated that the secondary structure of a miRNA precursor is similar to its neighboring miRNAs located in the same cluster, when compared to the sequences outside the clusters. Using this property, we designed a 2-step approach to filter the false positives resulted from a miRNA software tool and successfully raised the specificity by 10% while keeping a reasonably high sensitivity.