基于前馈人工神经网络的miRNA预测

摘要

microRNA( miRNA)是一类长度约为20～24个核苷酸保守的非编码小分子RNA,如何能准确预测miRNA一直是生物信息学的难点之一.文中提出一种新的预测方法-粒子群优化的前馈人工神经网络预测miRNA,从331(阴性数据168,阳性数据163)个样本组成的数据集中提取每个样本的36维特征向量训练人工神经网络模型,并用训练好的模型对不同的测试集进行测试,结果表明这种方法平均预测精度达到91.0％,高于传统的SVM预测方法,从而为miRNA预测提供了一个新的研究方向.%microRNA(miRNA} is a class of 20 -24 long nucleotides conserved non-coding small RNA. How to predict miRNA accurately is one of the difficulties in bioinformatics. A new predicting method has been proposed in this paper,that is,particle swarm optimized feedforward artificial neural network. Use 36 feature extracted from (he date set comprised of 331 samples to train the neural network model,which used to test new data-sets get a prediction accuracy up to 91.0% .This indicates that the model can be used as a new direction to predict miRNA.