Predicting the Protein-DNA Binding Residue Using Protein Language Model with Residual based CNN module

Numerous life-sustaining processes such as transcription, replication, and splicing, regulate the biological systems in a complex way. Therefore, it is important to recognize the pivotal role of crucial protein and DNA interactions for understand the above life sustaining processes. Hence, Protein-DNA binding residues is essential to identify as mentioned above. However, the classical experimental methods are tedious and labor-intensive. The recent research address these challenges and focused on predicting binding residues from sequence data. This present model utilizes a protein language model to transform protein sequence data into vector formats, enabling efficient computation. The model comprises three main components: an embedding layer that provide conversion of protein sequences into continuous vectors in the form of feature embedding using Protein-Bert model and a four-layer residual based convolutional neural network (CNN) trained on these vectors for prediction. The Binary Cross Entropy and contrastive loss function is used for loss computation. The same model, then reformatted, could be used to predict binding residues of RNA and antibodies. the present model achieves 0.884 AUC score and 0.795 specificity as compare to previous models.