Support Vector Machines for Predicting microRNA Hairpins

microRNAs (miRNAs) are 20-22 nt noncoding RNAs which are rapidly emerging as crucial regulators of gene expression in plants and animals. Identification of the hairpins which yield mature miRNAs is the first and most challenging step in miRNA gene prediction. We believe this step can best be achieved with biologically motivated feature design and classification techniques which account for the dependencies inherent in any set of hairpin features. We present DIANA-microH, a tool for predicting microRNA hairpins with high specificity and sensitivity. DIANA-microH implements a Support Vector Machine classifier trained on a set of structural and evolutionary features characteristic of miRNA hairpins. DIANA-microH introduces a unique structural feature motivated by a consideration of how enzymatic cleavage occurs. On test data, the SVM classifier achieved an accuracy of 98.6%. DIANA-microH is applied to chromosome 21 to provide a set of highly probable miRNA hairpins for future laboratory testing.