im7G-DCT: A two-branch strategy model based on improved DenseNet and transformer for m7G site prediction

Abstract

N-7 methylguanosine (m7G) is an important RNA modification that plays a key role in regulating gene expression and cellular physiological functions. Medical research has shown that m7G is closely associated with the development of a variety of diseases, including cancer, neurodegenerative diseases and viral infections. Therefore, accurate identification of m7G sites in mRNA is important for clinical applications and development of therapeutic strategies. With the rapid development of computational methods, deep learning prediction models are widely used in the field of m7G site. We developed a model called im7G-DCT based on the improved Densely Connected Convolutional Network and Transformer, which employs a two-branching strategy to extract local and global features from the original feature code in parallel. We found that this innovative strategy can mine the potential feature information of m7G locus sequences in a deeper way, which enhances the richness of features and improves the accuracy of prediction. As an innovative deep learning method, the im7G-DCT model shows important research value in the field of bioinformatics and has broad application prospects. In the results of independent test experiments, the im7G-DCT model achieved 81.68 %, 90.52 %, 86.10 %, and 72.48 % in sensitivity, specificity, accuracy, and Matthews' correlation coefficient, respectively. The im7G-DCT model performs well in all evaluation metrics and it significantly outperforms other existing prediction models.