Prediction of pre-miRNA with multiple stem-loops using pruning algorithm

Abstract

In addition to experimental identification of pre-miRNAs, the computational prediction method is also becoming a hot research spot. Most existing prediction methods are usually excluding those pre-miRNAs with multiple loops. But as more and more miRNA have been identified, quite a number of miRNA precursor with multiple loops have been found. Therefore, determining how to effectively identify pre-miRNAs with multiple loops from the control dataset with multiple loops is an imperative problem. In this work, a pruning algorithm is presented to identify the main branch from the multiple stem-loops of pre-miRNA. A stack algorithm is employed to describe the secondary structure of pre-miRNA in four different patterns, and a recursive algorithm is employed to split the multiple stem-loops of pre-miRNA into several small branches, and to identify its main branch. Statistic results indicate that the information of the main branch can be represented as the whole sequence of pre-miRNA. Some features of main branch are extracted to describe pre-miRNA intrinsic features, and SVM classifier is implemented to recognize real pre-miRNA with multiple stem-loops. Based on training and testing on dataset from miRBase12.0, SVM classifier achieves sensitivity of 75.76% on RM-POS and specificity of 98.12% on RM-CDS, and specificity of 91.28% on RM-NCR. The obtained results indicated that the information of main branch after pruning can represent intrinsic features of pre-miRNA with multiple stem-loops. The proposed method in this work provides a powerful predicting method to recognize the real pre-miRNA with multiple stem-loops.