Dual-Branch Contrastive Network with Deep Separable Convolution for Enhanced 6mA Site Identification

DNA N6-methyladenine (6mA) is a pivotal DNA modification integral to various biological processes, yet its exact regulatory role in eukaryotes is still unclear and controversial due to its sparsity, limitations in detection technologies, and complex regulatory mechanisms. In this study, we develop an innovative deep learning-based model for enhancing the prediction of 6mA sites, termed DS6mA, which uses a dual-branch contrastive network with deep separable convolution to extract the key position information from DNA sequences. First, the DNA sequence is encoded into feature vectors using a one-hot encoding method; Then, dual-branch networks with identical structures are formed and trained collaboratively using random paired samples to enhance the diversity of training data and improve the generalization ability of the model. Second, the features are input into the deep separable convolution, where residual connection is introduced through pointwise convolutions to enhance the expressive power of the feature vectors. Finally, the obtained features are fed into a fully connected neural network for the ultimate prediction. To effectively evaluate the performance of the model, we expanded the scope of the data sets examined in prior research by including 11 different comprehensive benchmark data sets, achieving favorable results. In summary, the proposed DS6mA method can effectively predict 6mA sites and has promising potential for future applications.