SeqDA-HLA: Language Model and Dual Attention-Based Network to Predict Peptide-HLA Class I Binding.

Area of Science:

• Immunology
• Computational Biology
• Bioinformatics

Background:

• Accurate prediction of peptide-HLA class I binding is essential for developing immunotherapies and vaccines.
• Current prediction methods face challenges in capturing complex peptide-HLA interactions.
• Diverse HLA alleles and peptide lengths require robust and generalizable prediction models.

Purpose of the Study:

• To introduce SeqDA-HLA, a pan-specific prediction model for peptide-HLA class I binding.
• To improve the accuracy and generalizability of peptide-HLA binding predictions.
• To provide an interpretable tool for identifying key binding features and motifs.

Main Methods:

• Utilized ELMo (Embeddings from Language Models) for rich contextual feature extraction.
• Implemented a dual attention mechanism, including self-aligned cross-attention and self-attention.
• Evaluated SeqDA-HLA against 14 state-of-the-art methods on multiple benchmark datasets.

Main Results:

• SeqDA-HLA achieved superior performance, with AUC up to 0.9856 and accuracy up to 0.9408.
• Demonstrated robust performance across various peptide lengths (8-14) and HLA alleles.
• Provided interpretable insights into essential anchor residues and binding motifs, aligning with biological data.

Conclusions:

• SeqDA-HLA is a powerful and interpretable tool for peptide-HLA binding prediction.
• The model shows significant potential for applications in epitope-based vaccine design and precision immunotherapy.
• Fine-tuning SeqDA-HLA on specific datasets, like Influenza, demonstrated its practical utility in predicting mutation-induced binding changes.