基于主成分分析-神经网络的非编码RNA预测

预测非编码RNA对认识其调控功能具有重要意义。选择单核苷酸和二核苷酸出现频率作为神经网络分类特征,运用主成分分析方法降低输入数据的维数,去除数据间的相关性,采用Levenberg-Marquardt算法改善网络训练速度。对数据集的测试结果表明,此方法对细菌混合ncRNA的预测精度达到81.3%,对原核生物tRNA的预测精度达到93.3%,表明该方法能有效预测原核生物ncRNA。预测结果还发现两种古细菌的ORF序列在序列特征上与其它细菌和古细菌存在差别。

Prediction of Non-coding RNA based on Neural Network with Principal Component Analysis

Prediction of Non-coding RNA(ncRNA) is important for exposing regulation functions of them.Appearance frequencies of single nucleotide and 2-nucleotides were chosen as characteristics inputs of a BP neural network.Before inputting to the model,dimensions of data were reduced,as well as the correlation being eliminated with the most information remained by principal component analysis(PCA).Levenberg-Marquardt algorithm was utilized to train the model.Results from tests showed that prediction accuracy was 81.3% on mixed bacteric ncRNAs dataset and 93.3% on prokaryotic tRNA dataset.The results indicate the model is adoptable for prokaryotic ncRNAs prediction.Different characteristics between ORFs from two kinds of archaebacteria and ORFs from other bacteria and archaebacteria were also discovered.