Structure Prediction

Our lab has developed a fine-tuned AlphaFold 2 model on peptide/HLA-I 9mer structures from HLA3DB. We have also used these models to predict conformational conservation in peptide backbones of pHLA complexes with PepPred.

This work, as well as our previous models, is available at these locations:

Benchmark results

We benchmarked our latest fine-tuned model (AFFT-HLA3DBv2) against current state-of-the-art prediction models as well as our own previous work. We compute pass rate and accuracy using D-Score cutoffs and giving emphasis to the conformational accuracy of the peptide backbone.

The advantage of AFFT-HLA3DBv2 over existing models is greatest in the non-A02 and non-∆7-1 cases, which accounts for its higher overall accuracy.

D-score accuracy by model across backbone subsets

References

Motmaen et al. Peptide-binding specificity prediction using fine-tuned protein structure prediction networks. Proc Natl Acad Sci USA. 2023 21 Feb. doi: 10.1073/pnas.2216697120

Gupta et al. HLA3DB: comprehensive annotation of peptide/HLA complexes enables blind structure prediction of T cell epitopes. Nat Commun 2023 10 Oct. doi: 10.1038/s41467-023-42163-z

Blackson et al. A generalizable system for antigenic peptide targeting across HLA-I allotypes. bioRxiv 2026.05.21.726655. doi: 10.64898/2026.05.21.726655