SVM-based miRNA: MiRNA∗ duplex prediction

We address the problem of predicting the miRNA: miRNA* duplex stemming from a microRNA (miRNA) hairpin precursor and we present a SVM-based methodology to address it. Predicting the miRNA: miRNA* duplex is a first step towards identifying the mature miRNA, suggesting possible miRNA targets and ultimately, reducing experimentation effort, time, and cost. We measure the error in terms of the absolute difference of the true and predicted location of all of the four ends of the duplex and/or of each end separately. Our mean absolute error over all ends is 1.61 ± 2.24 nts as measured on a hold-out set of 220 miRNA hairpin precursor sequences. In addition, our tool precisely predicts (with 0 nt deviation) the starting position for 57% and 52% of the miRNAs in the 5' and 3' strands of the same dataset, significantly outperforming the state-of-the-art tool MaturePred which achieves 18% and 12%, respectively, on the same task. Overall, our method accurately identifies not only the starting nucleotide of novel miRNA: miRNA* duplexes - and thus individual miRNAs- but also their length, while outperforming the current state-of-the-art tool.