Gaussian-Based Dilated 1-D CNN for Classifying B-Cell Epitopes in Zika and Dengue Protein Sequences

Abstract

The main goal of designing peptide vaccines, conducting immunodiagnosis, and producing antibodies is to accurately identify linear B-cell epitopes. However, experimental analysis to determine these epitopes is costly. This study focuses on developing a Gaussian-based dilated 1-D CNN model for classifying epitopes and non-epitopes in protein sequences related to Zika and Dengue viruses. The Immune Epitope Database (IEDB) was used, containing a total of 1741 and 7020 linear B-cell epitopes for Zika and Dengue viruses, respectively. Physicochemical features of the protein sequences dataset were extracted using the Gaussian distribution to extract optimal features based on feature probability distribution. The proposed model achieved an accuracy score of 83.00% and 85.00%, precision of 87.00%, recall of 83.00% and 85.00%, and an F1-score of 84.00% and 86.00% over the Zika and Dengue datasets. The suggested model outperforms existing methods, demonstrating the potential of deep learning approaches in bioinformatics for enhancing epitope prediction in viruses, with implications for drug discovery and vaccine development.