DeepRice6mA: A convolutional neural network approach for 6mA site prediction in the rice Genome

Abstract

As one of the most critical post-replication modifications, N6-methylation (6mA) at adenine residue plays an important role in a variety of biological functions. Existing computational methods for identifying 6mA sites across large genomic regions tend to fall short in either accuracy or computational efficiency. To address this, we introduce DeepRice6mA, a sophisticated comprehensive predictive tool for identifying rice 6mA sites, using a deep learning approach that incorporates ensemble strategies from one-hot encoding and 3-kmer feature embedding. The proposed model, labeled DeepRice6mA, reaches state-of-the-art results compared to current approaches, with 10-fold cross-validation scores of 98% for accuracy, 98% for sensitivity, 98% for specificity, a Matthew’s correlation coefficient (MCC) of 0.96, and an area under the receiver operating characteristic curve (AUC) of 0.99. We anticipate that DeepRice6mA will significantly enhance our understanding of DNA methylation and its implications for biological processes and disease states.