SCBi-GRUCapformer: prediction of multi-class nucleic acid-binding proteins (NABPs) using squash conventional BiGRU capsule transformer with feature selection

The precise identification of binding proteins is necessary, but it's more critical. According to the low prediction accuracy, structural similarities, and a significant cross-prediction rate, RBP predictors usually anticipate the DBPs as the RBPs, and DBP predictors commonly estimate the RBPs as the DBPs. To address these issues, this research proposed a novel binding protein prediction (BPP) technique, which can be termed as Squash convolutional BiGRU capsule transformer (SCBi-GRUCapformer) model. In this research, the four phases, such as pre-processing, feature extraction, feature selection, and binding protein prediction (BPP), can be utilized in order to predict the protein sequence accurately. The redundant sequences of lengths fewer than 40 are eliminated from the protein sequence through the pre-processing phase. During the feature extraction step, the Python protlearn module was utilized to extract features from protein sequences, which include atomic composition (ATC), dipeptide composition (DPC), amino acid composition (AAC), and physicochemical properties (PCP). Furthermore, during the feature selection stage, insignificant attributes are deleted, and a Learning Alpine Skiing Imitation Optimization (LASIO) technique is used to select the best features. Finally, the SCBi-GRUCapformer model is utilized to predict the BPP in four different classes, namely DBPs, RBPs, DRBPs, and NNABP as Class 0, 1, 2, and 3, respectively. In the simulation scenario, several performance measures are analyzed, and they attain an outcome of 97.84% accuracy, 95.27% precision, 95.14% recall, and 95.2% f-measure, respectively. In addition to this, the ablation study and the cross-validation analysis are also conducted to show the efficacy of the proposed model.