A Unified Framework for TCR-pMHC Structural Model Assessment
1. This study introduces a novel framework for assessing the quality of TCR-pMHC structural models without requiring experimental reference structures. The framework leverages multiple modeling and interface confidence metrics integrated into explainable random forest classifiers, trained on 1160 models of 232 experimentally determined PDB TCR-pMHC class I complexes. This approach reliably distinguishes between low-, acceptable-, medium-, and high-quality models, outperforming traditional metric thresholds.
2. The study benchmarks five state-of-the-art protein modeling tools (AlphaFold3, Boltz-2, Chai-1, tFold-TCR, and TCRmodel2) on a set of 20 experimentally resolved TCR-pMHC class I structures. AlphaFold3 demonstrated superior performance, with 90% of models classified as acceptable or better, including 20% high-quality models. This highlights the potential of AlphaFold3 for generating reliable TCR-pMHC models.
3. The researchers expanded the structural landscape of TCR-pMHC class I complexes by generating a large synthetic dataset using AlphaFold3. This dataset includes 33,820 unique complexes (169,100 models) from VDJdb-annotated sequences, representing a >70-fold increase in available structures. This expansion provides a valuable resource for downstream applications such as epitope mapping and predictive modeling.
4. The quality tier framework was applied to filter and validate TCR-pMHC interactions in sequence databases like VDJdb. Higher-quality models were significantly enriched for biologically validated interactions, with high-quality models achieving the highest enrichment. This suggests that model quality can be used as a filtering criterion to identify false positives in sequence databases.
5. The study also demonstrates that high-quality structural models enhance the predictive performance of TCR-pMHC pairing algorithms. Using the IMMREP23 dataset, the researchers showed that higher-quality models are enriched for positive TCR-pMHC pairs and improve the accuracy of pairing predictions. This highlights the potential of the framework for improving TCR-based immunotherapies.
📜Paper:
biorxiv.org/content/10.1101/…
#TCR #pMHC #StructuralModeling #AlphaFold3 #Immunology #Bioinformatics