Attention Integrated Residue CNN for Classification of DNA-Binding and RNA-Binding Proteins

The classification of DNA-binding proteins (DBPs) and RNA-binding proteins (RBPs) is essential for understanding molecular interactions and regulatory functions. However, the existing computational approaches face difficulties in differentiating between DBPs and RBPs resulting in cross-prediction errors. To mitigate this issue, SERCNN as a multilabel classification model, which integrates deep neural networks with residual connections and squeeze-and-excitation attention mechanisms, is developed in this paper. This approach provides improved performance in predicting proteins that bind with DNA, RNA, and the proteins that bind to both DNA and RNA (DRBPs). On the TEST474 dataset, the SERCNN achieved an AUC value of 0.95 with a 1-AURC value of 0.91 for the prediction of DBPs. For RBP prediction an AUC value of 0.92 with a 1-AURC value of 0.90, is obtained. For the PDB255 dataset, the proposed model achieved an AUC of 0.80 and a 1-AURC of 0.85 for DBPs and an AUC of 0.80 and a 1-AURC of 0.81 for RBPs. Improved prediction performance has also been obtained on the DRBP206 dataset. Therefore, the prediction results obtained on the test dataset suggest that the proposed model can employed as a useful tool for DBP and RBP prediction tasks.