HN-CNN: A Heterogeneous Network Based on Convolutional Neural Network for m(7) G Site Disease Association Prediction

N(7)-methylguanosine (m(7)G) is a typical positively charged RNA modification, playing a vital role in transcriptional regulation. m(7)G can affect the biological processes of mRNA and tRNA and has associations with multiple diseases including cancers. Wet-lab experiments are cost and time ineffective for the identification of disease-related m(7)G sites. Thus, a heterogeneous network method based on Convolutional Neural Networks (HN-CNN) has been proposed to predict unknown associations between m(7)G sites and diseases. HN-CNN constructs a heterogeneous network with m(7)G site similarity, disease similarity, and disease-associated m(7)G sites to formulate features for m(7)G site-disease pairs. Next, a convolutional neural network (CNN) obtains multidimensional and irrelevant features prominently. Finally, XGBoost is adopted to predict the association between m(7)G sites and diseases. The performance of HN-CNN is compared with Naive Bayes (NB), Random Forest (RF), Support Vector Machine (SVM), as well as Gradient Boosting Decision Tree (GBDT) through 10-fold cross-validation. The average AUC of HN-CNN is 0.827, which is superior to others.