MHCmet: A neural network based epitope prediction tool for orthohantaviruses

Area of Science:

• Immunoinformatics
• Computational Biology
• Vaccine Development

Background:

• Identifying immunogenic peptides is crucial for developing diagnostics, therapeutics, and vaccines.
• Traditional methods are inefficient and lack scalability for large proteomes.
• Computational tools are essential for identifying immunogenic fragments in viral proteomes.

Purpose of the Study:

• To develop MHCmet, a deep learning tool for identifying peptide fragments that bind human MHC class I receptors.
• To improve the accuracy and efficiency of immunogenic peptide prediction.
• To provide a novel computational approach for vaccine design.

Main Methods:

• Developed MHCmet, a deep learning tool integrating CNN-bidirectional LSTM layers.
• Implemented an allele-specific encoding strategy, dynamically selecting One-hot, BLOSUM62, or NLF.
• Utilized multi-head self-attention and bidirectional LSTM layers to capture peptide-MHC binding groove diversity.

Main Results:

• MHCmet achieved a median AUC of 0.91, outperforming NetMHCpan (0.89) and MHCflurry (0.85).
• The tool successfully predicted over 40 peptide fragments binding to MHC class I receptors for Dobrava Orthohantavirus (DOBV).
• MHCmet accepts FASTA sequences and provides detailed epitope prediction data.

Conclusions:

• MHCmet offers a significant advancement in predicting immunogenic peptide fragments.
• The tool's allele-specific encoding enhances prediction accuracy for diverse MHC alleles.
• Predicted epitopes require experimental validation for translational applications in diagnostics and therapeutics.