Computational classification of microRNAs in next-generation sequencing data

Abstract

MicroRNAs (miRNAs) have been shown to play an important regulatory role in plants and animals. A large number of known and novel miRNAs can be uncovered from next-generation sequencing (NGS) experiments that measure the complement of a given cell’s small RNAs under various conditions. Here, we present an algorithm based on radial basis functions for the identification of potential miRNA precursor structures. Computationally assessing features of known human miRNA precursors, such as structural linearity, normalized minimum folding energy, and nucleotide pairing frequencies, this model robustly differentiates between miRNAs and other types of non-coding RNAs. Without relying on cross species conservation, the method also identifies non-conserved precursors and achieves high sensitivity. The presented method can be used routinely for the identification of known and novel miRNAs present in NGS experiments.