M6A-BERT-Stacking: A Tissue-Specific Predictor for Identifying RNA N6-Methyladenosine Sites Based on BERT and Stacking Strategy

As the most abundant RNA methylation modification, N6-methyladenosine (m6A) could regulate asymmetric and symmetric division of hematopoietic stem cells play an important role in various diseases. Therefore, precise identification m6A sites around genomes different species is a critical step to further revealing their biological functions influence on these However, traditional wet-lab experimental methods for identifying are often laborious expensive. In this study, we proposed ensemble deep learning model called m6A-BERT-Stacking, powerful predictor detection tissues three species. First, utilized two encoding methods, i.e., di ribonucleotide index (DiNUCindex_RNA) k-mer word segmentation, extract sequence features. Second, matrices together with original sequences were respectively input into models parallel train sub-models, namely residual networks convolutional block attention module (Resnet-CBAM), bidirectional long short-term memory (BiLSTM-Attention), pre-trained encoder representations from transformers DNA-language (DNABERT). Finally, outputs all sub-models ensembled based stacking strategy obtain final prediction through fully connected layer. The results demonstrated that m6A-BERT-Stacking outperformed existing same independent datasets.