MHCSeqNet2-improved peptide-class I MHC binding prediction for alleles with low data

Area of Science:

• Immunology
• Computational Biology
• Bioinformatics

Background:

• Peptide-MHC class I binding is vital for immune recognition of infected or cancerous cells.
• Accurate prediction of peptide-MHC binding is essential for developing effective peptide-based cancer vaccines.
• Limited experimental data for many MHC alleles hinders the accuracy of existing prediction models.

Purpose of the Study:

• To present an improved peptide-MHC binding prediction model, MHCSeqNet2.
• To enhance model generalizability for MHC alleles with limited available data.
• To improve the prioritization of T cell epitopes for personalized cancer therapies.

Main Methods:

• Utilized sub-word-level peptide features.
• Incorporated 3D structure embeddings for MHC alleles.
• Employed an expanded training dataset for improved generalizability.
• Visualized MHC allele embeddings to confirm grouping by binding specificity.

Main Results:

• MHCSeqNet2 demonstrates improved generalizability on MHC alleles with scarce data.
• Visualization confirmed the model's ability to cluster alleles with similar binding specificities.
• External evaluation indicated enhanced prioritization of T cell epitopes for data-limited MHC alleles.

Conclusions:

• MHCSeqNet2 offers a significant advancement in predicting peptide-MHC binding, particularly for underrepresented MHC alleles.
• The model's architecture and training approach contribute to better performance and broader applicability in cancer immunotherapy.
• The improved prediction accuracy facilitates the selection of optimal peptides for patient-specific vaccines.