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

Abstract

The precise prediction of major histocompatibility complex (MHC)-peptide complex structures is pivotal for understanding cellular immune responses and advancing vaccine design. In this study, we enhanced AlphaFold's capabilities by fine-tuning it with a specialized dataset consisting of exclusively high-resolution class I MHC-peptide crystal structures. This tailored approach aimed to address the generalist nature of AlphaFold's original training, which, while broad-ranging, lacked the granularity necessary for the high-precision demands of class I MHC-peptide interaction prediction. A comparative analysis was conducted against the homology-modeling-based method Pandora as well as the AlphaFold multimer model. Our results demonstrate that our fine-tuned model outperforms others in terms of root-mean-square deviation (median value for Cα atoms for peptides is 0.66 Å) and also provides enhanced predicted local distance difference test scores, offering a more reliable assessment of the predicted structures. These advances have substantial implications for computational immunology, potentially accelerating the development of novel therapeutics and vaccines by providing a more precise computational lens through which to view MHC-peptide interactions.

Figure 1

Comparative analysis of prediction accuracy for class I MHC-peptide complexes using RMSD calculated for Cα atoms of peptides: performance of AlphaFold, Pandora, and MHC-Fine.

Figure 2

Comparative visualization of class I MHC-peptide complex prediction accuracy, reporting MHC-peptide RMSD calculated for Cα atoms. True peptide structure is shown in red, Pandora model in orange, and MHC-Fine in blue. (a) PDB: 6vb3: high precision of MHC-Fine. MHC-peptide RMSD for Cα atoms is 0.25 Å for MHC-Fine and 1.44 Å for Pandora. (b) PDB: 7n2o: moderate accuracy for both models. MHC-peptide RMSD for Cα atoms is 1.23 Å for MHC-Fine and 1.19 Å for Pandora. (c) PDB: 7mj7: significant deviations in predictions for both models. MHC-peptide RMSD for Cα atoms is 3.73 Å for MHC-Fine and 4.30 Å for Pandora. To see this figure in color, go online.

Figure 3

Error distribution for plDDT values of peptide residues. The red distribution represents the original AlphaFold model with a standard deviation of 9.2, whereas the blue distribution highlights our MHC-Fine model, which achieves a reduced standard deviation of 4.6, indicating a more precise confidence in structural predictions for peptide residues.

Figure 4

MHC-Fine confidence prediction. Samples are grouped by mean plDDT values for peptide residues. The chart displays the distribution of MHC-peptide RMSD alongside the percentage of samples within each specific plDDT range.