| Introduction: MHC class II molecules play a critical role in adaptive immunity by presenting antigenic peptides to CD4⁺ T cells, a process fundamental to vaccine design, immunotherapy, and disease diagnostics. However, conventional epitope prediction methods—primarily based on sequence alignment or traditional machine learning—struggle to capture the inherent complexity and diversity of protein sequences, limiting their predictive accuracy and generalizability. Methods: We introduce MHC-II-EpiPred, a novel deep learning-based framework for MHC class II epitope prediction, leveraging the ESM-2 protein language model. The model is fine-tuned on a curated dataset of experimentally validated epitopes from the Immune Epitope Database (IEDB) and employs a sliding window approach for systematic sequence scanning. To enhance computational efficiency and model robustness, we incorporate FP16 precision, dropout regularization, weight decay, early stopping, and low-rank adaptation (LoRA) during training. Results: MHC-II-EpiPred achieves state-of-the-art performance, with a training cross-entropy loss (CEL) of 0.1407 and an accuracy of 98.98%, while maintaining an evaluation CEL of 0.0836 with an accuracy of 97.03%. These results confirm the model's ability to effectively capture the complex sequence patterns associated with MHC class II binding epitopes. Conclusion: MHC-II-EpiPred provides a powerful and accessible tool for MHC class II epitope prediction. this framework is expected to facilitate advancements in vaccine development, immunotherapy, and immunological research. |