Deep Learning Prediction of Antigen–Epitope Immunogenicity for Peptide Vaccine Design

Background: Peptide-based vaccines rely on the identification of antigenic epitopes capable of inducing strong immune responses. However, predicting immunogenic peptide sequences remains challenging due to the complex interaction between amino-acid composition, structural features, and immune recognition mechanisms. Artificial intelligence approaches may improve epitope selection efficiency in early vaccine design.

Methods: A dataset containing 6,180 experimentally characterised peptide epitopes and non-epitope sequences was compiled from publicly accessible immunological databases. Each sequence was encoded using physicochemical descriptors including hydrophobicity index, amino acid polarity distribution, molecular flexibility, and predicted secondary structure probability. A convolutional neural network with three convolution layers and two dense layers was trained to classify peptide immunogenicity. Data were divided into training (70%), validation (15%), and independent test sets (15%). Model performance was evaluated using accuracy, F1-score, area under the receiver operating characteristic curve (AUC), precision, and recall.

Results: The trained model achieved an overall classification accuracy of 89.6% on the independent test dataset with an AUC of 0.93 and F1-score of 0.88. Precision and recall values were 0.90 and 0.86, respectively. Feature attribution analysis revealed that hydrophobic residue clustering and peptide flexibility were the most influential determinants of predicted immunogenicity. The model correctly identified 92% of high-immunogenic peptides within the validation dataset and reduced false-positive predictions by 18% compared with baseline logistic regression models.

Conclusion: Deep learning models incorporating physicochemical peptide descriptors can reliably predict antigenic epitopes with high immunogenic potential. Such computational tools may support the rational design of peptide-based vaccines and accelerate early-stage vaccine development workflows.

To be updated shortly..