Prediction of RNA-protein interactions by combining deep convolutional neural network with feature selection ensemble method

JOURNAL OF THEORETICAL BIOLOGY

卷:461

Article

Wang, Lei¹  Yan, Xin²  Liu, Meng-Lin¹  Song, Ke-Jian³  Sun, Xiao-Fei¹  Pan, Wen-Wen¹ 

[1] Zaozhuang Univ, Coll Informat Sci & Engn, Zaozhuang 277100, Shandong, Peoples R China

[2] Zaozhuang Univ, Sch Foreign Languages, Zaozhuang 277100, Shandong, Peoples R China

[3] JiangXi Univ Sci & Technol, Sch Informat Engn, Ganzhou 341000, Jiangxi, Peoples R China

关键词: RNA-protein interaction;    Convolution neural network;    Extreme learning machine;    Position-specific scoring matrix;   

DOI  :  10.1016/j.jtbi.2018.10.029

来源: Elsevier

【 摘 要 】

RNA-protein interaction (RPI) plays an important role in the basic cellular processes of organisms. Unfortunately, due to time and cost constraints, it is difficult for biological experiments to determine the relationship between RNA and protein to a large extent. So there is an urgent need for reliable computational methods to quickly and accurately predict RNA-protein interaction. In this study, we propose a novel computational method RPIFSE (predicting RPI with Feature Selection Ensemble method) based on RNA and protein sequence information to predict RPI. Firstly, RPIFSE disturbs the features extracted by the convolution neural network (CNN) and generates multiple data sets according to the weight of the feature, and then use extreme learning machine (ELM) classifier to classify these data sets. Finally, the results of each classifier are combined, and the highest score is chosen as the final prediction result by weighting voting method. In 5-fold cross-validation experiments, RPIFSE achieved 91.87%, 89.74%, 97.76% and 98.98% accuracy on RPI369, RPI2241, RPI488 and RPI1807 data sets, respectively. To further evaluate the performance of RPIFSE, we compare it with the state-of-the-art support vector machine (SVM) classifier and other exiting methods on those data sets. Furthermore, we also predicted the RPI on the independent data set NPInter2.0 and drew the network graph based on the prediction results. These promising comparison results demonstrated the effectiveness of RPIFSE and indicated that RPIFSE could be a useful tool for predicting RPI. (C) 2018 Elsevier Ltd. All rights reserved.

【 授权许可】

Free