A new method for identification of pre-microRNAs based on hybrid features

Abstract

<strong>Background:</strong>The identification of pre-microRNAs (precursor microRNAs) helps us<br/>to understand the regulatory mechanism of biological processes.<br/>Currently, machine learning is the most popular method for<br/>pre-microRNA identification. However, most methods mainly focus on<br/>secondary structure information of pre-microRNA, while ignoring<br/>sequence-order information and sequence evolution information.<strong>Results:</strong><br/>In this work, we use three different methods to<br/>extract features of the pre-microRNAs at different levels. We first<br/>extract features from PSI-BLAST profiles and Hilbert-Huang<br/>transform, which contain rich sequence evolution information and<br/>sequence-order information respectively. We then get properties of<br/>small molecular networks of pre-microRNAs, which contain refined<br/>secondary structure information. We extract 591 features in total.<br/>After extraction, we use support vector machine (SVM) as our<br/>classifier, and use the maximum relevance and minimum redundancy<br/>(mRMR) method for feature selection. Finally, we construct a new<br/>predictor <em>MicroRNA-NHPred</em> by using the optimal feature set.<br/>The performance of <em>MicroRNA-NHPred</em> is quite promising<br/>compared to other popular miRNA predictors. It achieves an accuracy<br/>of up to 94.83\%.<strong>Conclusions:</strong><br/>The higher prediction accuracy achieved by our proposed method is<br/>attributed to the design of a comprehensive feature set on the<br/>sequence and secondary structure, which are capable of<br/>characterizing the sequence evolution information and sequence-order<br/>information, and global and local information of pre-microRNAs<br/>secondary structure. Therefore, it is a valuable method to<br/>pre-microRNAs identification.