Structural quality-tier assessment for TCR-pMHC functional enrichment

Area of Science:

• Immunology and Structural Biology
• Computational Biology and Bioinformatics

Background:

• T cell receptor (TCR) recognition of peptide-MHC complexes (pMHCs) is fundamental to adaptive immunity and T cell function.
• Experimental structure determination of TCR-pMHC complexes is limited, hindering detailed analysis.
• Accurate computational modeling of TCR structures is challenging due to loop hypervariability and flexibility, with a lack of reliable quality assessment methods.

Purpose of the Study:

• To benchmark the performance of general-purpose and TCR-specific protein structure modeling algorithms for TCR-pMHC complexes.
• To develop and validate a machine learning (ML) framework for assessing the quality of TCR-pMHC structural models without experimental references.
• To enable prioritization of high-confidence TCR-pMHC interactions for immunological and therapeutic applications.

Main Methods:

• Recalculated all experimentally determined TCR-pMHC class I complexes using AlphaFold2.3-Multimer, AlphaFold3, Boltz-2, Chai-1, TCRmodel2, tFold-TCR, and TCRdock.
• Developed a random forest classifier trained on confidence metrics (pLDDT, ipTM, ipSAE, iPAE, iPDE, pDockQv1-2) from 1325 modelled structures of 265 experimentally determined complexes.
• Validated the ML framework on two independent datasets (VDJdb and IMMREP23) comprising over 4,000 AlphaFold3-modelled TCR-pMHC complexes.

Main Results:

• AlphaFold3 demonstrated superior performance in modeling TCR-pMHC class I complexes (mean TCR-iRMSD = 3.59 Å, DockQ = 0.54) compared to other algorithms.
• The ML quality assessment framework reliably stratified structural models into low, acceptable, medium, and high-quality tiers, outperforming single confidence metrics.
• The framework successfully profiled VDJdb to reduce false-positive interactions and enriched for biologically validated interactions in the IMMREP23 dataset.

Conclusions:

• AlphaFold3 is a leading tool for TCR-pMHC structural modeling, and the developed ML framework offers a scalable and interpretable method for quality assessment.
• This quality-tier framework enhances the reliability of computational TCR-pMHC models, aiding in the prioritization of high-confidence interactions.
• The approach has direct translational applications in T-cell immunology, TCR-based immunotherapies, and understanding antigen specificity.