Transfer learning improves pMHC kinetic stability and immunogenicity predictions.

Area of Science:

• Immunology
• Computational Biology
• Bioinformatics

Background:

• Cellular immune response involves peptide-Major Histocompatibility Complex (pMHC) binding, presentation, and T-Cell Receptor recognition.
• Accurate prediction of peptide targets is vital for peptide-based vaccines and T-cell immunotherapies.
• Machine learning (ML) excels at pMHC binding prediction, but accuracy for kinetic stability and immunogenicity is limited by data scarcity.

Purpose of the Study:

• To enhance the prediction accuracy of peptide-MHC kinetic stability and immunogenicity.
• To leverage transfer learning techniques to overcome data limitations in stability and immunogenicity prediction.
• To develop data-driven tools that improve the design of peptide-based vaccines and immunotherapies.

Main Methods:

• Utilized transfer learning by applying knowledge from large peptide-MHC binding affinity and mass spectrometry datasets.
• Developed two models, TLStab and TLImm, to predict peptide stability and immunogenicity, respectively.
• Evaluated model performance against state-of-the-art approaches on independent test sets.

Main Results:

• The developed models, TLStab and TLImm, achieved performance comparable to or exceeding existing methods.
• Demonstrated the effectiveness of transfer learning in improving predictions for peptide stability and immunogenicity.
• Validated the approach using diverse stability and immunogenicity test datasets.

Conclusions:

• Transfer learning is a promising strategy for improving predictions in areas with limited specialized datasets, such as peptide stability and immunogenicity.
• The developed models offer enhanced predictive capabilities for critical steps in the cellular immune response.
• This work facilitates the advancement of peptide-based vaccine and immunotherapy development through improved computational tools.