AMmRNA_BBC

A novel multi-stage AMmRNA-BBC model is introduced with a novel approach to biological data and their combination. In the first stage, a feature extraction method based on multimodal and innovative features such as forward and reverse alignment-based nucleotide pairs (NucP), mRNA subcellular abundance-position rate in the k-nucleotide subrange (kNF-PR), and protein-protein interaction (PPI) data is proposed. Considering the challenge of the large number of extracted features, a hybrid deep network based on the attention mechanism based on Self-Attention and Cross-Attention is proposed to map features into a separable space and extract relationships between effective features and in-situ patterns. Imbalance of samples in biological data based on mRNA location positions is one of the main challenges in this field. In this regard, in the next step, a combined method of Random Oversampler and BBC classifier is presented for accurate and robust classification against data imbalance. The proposed method is evaluated on Cefra-Seq and RNALocate datasets and the results show an improvement in the performance of the proposed method compared to state-of-the-art methods.