MHC-Fine: Fine-tuned AlphaFold for precise MHC-peptide complex prediction

Area of Science:

• Computational immunology
• Structural biology
• Vaccine development

Background:

• Accurate prediction of major histocompatibility complex (MHC)-peptide complex structures is crucial for understanding T-cell mediated immunity.
• Existing generalist models like AlphaFold may lack the specificity required for high-precision MHC-peptide interaction prediction.

Purpose of the Study:

• To enhance AlphaFold's predictive accuracy for class I MHC-peptide complex structures.
• To develop a specialized model for high-resolution MHC-peptide structure prediction.

Main Methods:

• Fine-tuning AlphaFold using a curated dataset of high-resolution class I MHC-peptide crystal structures.
• Comparative performance analysis against Pandora (homology modeling) and AlphaFold multimer.

Main Results:

• The fine-tuned model demonstrated superior performance, evidenced by a lower median root-mean-square deviation (0.66 Å for peptide Cα atoms).
• Improved predicted local distance difference test scores indicate more reliable structural predictions.
• Outperformed both Pandora and AlphaFold multimer in accuracy.

Conclusions:

• Specialized fine-tuning significantly improves AlphaFold's capability for precise MHC-peptide structure prediction.
• This advancement offers a more reliable computational tool for drug discovery and vaccine design in immunology.
• Enhanced MHC-peptide structure prediction accelerates research in computational immunology and therapeutic development.