ConvNeXt-MHC: improving MHC-peptide affinity prediction by structure-derived degenerate coding and the ConvNeXt model

Area of Science:

• Immunoinformatics
• Computational Biology
• Machine Learning in Immunology

Background:

• Peptide binding to Major Histocompatibility Complex (MHC) proteins is fundamental for T-cell recognition and immune response specificity.
• Experimental validation of peptide-MHC binding is resource-intensive, necessitating accurate computational prediction methods.
• Accurate prediction is vital for applications like cancer immunotherapy, particularly for neoantigen identification.

Purpose of the Study:

• To develop an advanced computational method for predicting MHC-I-peptide binding affinity.
• To improve the accuracy and efficiency of predicting peptides relevant to T-cell recognition.
• To provide a tool that aids in the discovery of neoantigens for cancer immunotherapy.

Main Methods:

• Developed ConvNeXt-MHC, a deep learning framework utilizing the ConvNeXt architecture.
• Integrated transfer learning and semi-supervised learning techniques.
• Employed a degenerate encoding approach to enhance panspecific prediction capabilities.

Main Results:

• ConvNeXt-MHC demonstrated superior accuracy in predicting MHC-I-peptide binding affinity compared to state-of-the-art methods.
• Comprehensive benchmark results validated the method's performance.
• The method shows promise for advancing immunoinformatics research.

Conclusions:

• ConvNeXt-MHC offers a significant advancement in predicting MHC-I-peptide binding affinity.
• The developed method is expected to facilitate new discoveries in immunoinformatics and cancer immunotherapy.
• A user-friendly web application is available for accessing the tool and data.