Aug 2026· Proceedings of the 32nd ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.2· pp. 12468-12479· 0 citations· 10 references
Abstract
Designing functional T cell receptors (TCRs) for a given peptide presented by MHC (pMHC) is an emerging yet highly challenging problem in computational immunology. While recent approaches have achieved initial progress, they face two major limitations: (1) the lack of structural information from TCR–pMHC complexes in the design process, and (2) the restricted generalization ability of current sequence–structure co-design models, which rely only on paired sequence–structure data and fail to leverage the vast amount of available sequence-pairing information. To address these challenges, we introduce TCRTSdesign, a framework that concurrently generates novel TCR sequences with specific binding capabilities to target pMHC molecules and predicts the full-atom structures of the TCR-pMHC complex, while optimizing their binding affinity. Our method integrates large-scale paired sequence data for pretraining a sequence generation model, and further refines the design through a structure-aware student model guided by the teacher via knowledge distillation. Extensive experiments demonstrate that TCRTSdesign significantly outperforms existing baselines in both sequence recovery and structural fidelity, offering a promising computational method for TCR engineering.
Adaptive immunity relies on T-cell receptor (TCR) recognition of peptides presented by the major histocompatibility complex (pMHC). Accurate prediction of TCR:pMHC binding pairs from sequence data remains a longstanding challenge in computational immunology, limiting the development of precision immunotherapies like cancer vaccines and adoptive cell therapies. Here, we present enFoldX (ensemble of Folded compleXes), a structure-based approach leveraging biophysical characterization of AlphaFold3-generated ensembles to classify TCR:pMHC sequence pairs as cognate versus non-cognate. Unlike previous methods reliant on only sequence data or a single, static predicted structure, enFoldX extracts features from an entire generated ensemble with a custom focus on the biophysical binding interface. Our model distinguishes T cell reactivity between peptides differing by a single amino acid substitution, the resolution required for cancer neoantigens, and generalizes to unseen peptides, MHCs, and TCRs, a major objective for artificial intelligence (AI) in immunology. Our performance on these crucial tasks demonstrates that diverse, structural sampling of biophysical interactions over an ensemble is fundamental for accurate AI-driven binding predictions and offers lessons for efficient future data generation to improve models. Our findings therefore offer a scalable framework to accelerate therapeutic binder design, and we provide access to a publicly available code repository.
O. Lyudovyk, JA Levine, M. Pathil et al.· bioRxiv· 1 citation
T cell receptor (TCR) recognition of peptide-MHC complexes (pMHCs) is central to adaptive immunity. Structural insights into TCR-pMHC interactions are critical for understanding antigen specificity and T-cell function. However, progress remains limited by the scarcity of experimentally resolved structures (275 TCR-pMHC class I structures in the PDB, Jan 2026). Although protein structure modelling tools have advanced rapidly, accurate structural modelling of TCRs remains challenging due to CDR loop hypervariability and conformational flexibility. In addition, there is a lack of reliable quality assessment strategies that do not rely on comparisons with experimental references. To address this, we benchmarked four general-purpose (AlphaFold2.3-Multimer, AlphaFold3, Boltz-2, Chai-1) and three TCR-specific (TCRmodel2, tFold-TCR, TCRdock) protein modelling algorithms by recalculating all experimentally determined TCR-pMHC class I complexes in the PDB. AlphaFold3 demonstrated superior performance across metrics (mean TCR-iRMSD = 3.59 Å and DockQ = 0.54), whereas the other algorithms displayed lower accuracy. Built on AlphaFold3 structures, we present a scalable and interpretable ML framework for the quality assessment of TCR-pMHC structural models without matched experimental references. We trained a random forest classifier integrating multiple confidence metrics (pLDDT, ipTM, ipSAE, iPAE, iPDE, pDockQv1-2) derived from 1325 modelled structures of 265 experimentally determined PDB TCR-pMHC class I complexes. The classifier reliably stratifies structural models into low-, acceptable-, medium-, and high-quality tiers defined by comparisons to their experimental reference structures, outperforming single metrics. These quality-tier predictions further enable the prioritization of high-confidence TCR-pMHC interactions. This was demonstrated across two held-out datasets comprising a total of 4,090 AlphaFold3-modelled TCR-pMHC complexes (20,450 models, 5 models per complex): a re-evaluated set of TCR-pMHC class I complexes from VDJdb (n = 606) and the reference IMMREP23 dataset (n = 3,484). We profiled the first dataset to reduce false-positive TCR-pMHC interactions erroneously annotated in VDJdb, and the second to enrich for biologically validated TCR-pMHC interactions amongst higher-quality structural models over their synthetic negative counterparts. Altogether, our structural quality-tier framework provides a scalable and interpretable approach that complements structural modelling and functional analyses of TCR-pMHC class I complexes, with direct translational applications in T-cell immunology and TCR-based immunotherapies.
Alex Ascunce-París, Miguel Romero-Durana, Alfonso Valencia et al.· Frontiers in Immunology· 0 citations
Specific recognition between T-cell receptors (TCRs) and peptide-major histocompatibility complexes (pMHCs) is central to adaptive immunity, yet accurate prediction of TCR-pMHC specificity remains challenging. Existing models mainly rely on sequence features or isolated molecular structures, limiting their ability to capture interface-level determinants within the ternary recognition complex. Here, we constructed the multimodal TCR-pMHC ternary complex (MM-TCR) data set, integrating paired TCR-pMHC sequences, V/J gene annotations, and modeled TCR-pMHC complex structures refined by short molecular dynamics-based relaxation. Based on MM-TCR, we developed TCRspec, an interpretable multimodal framework combining sequence embeddings, gene-usage features, and complex-level structural representations. Under a stringent CD-HIT TCR-cluster-disjoint split, TCRspec achieved an average AUROC of 0.896 and AUPRC of 0.882 across seven antigen-specific test data sets, outperforming representative baseline models. Cross-validation and ablation analyses confirmed the contribution of ternary complex structural information and MD-refined structures. In independent OOD peptide-TCR systems, TCRspec retained discriminative performance and identified model-inferred peptide positions associated with TCR recognition, providing a structure-informed framework for TCR specificity prediction.
Jia Zou, Zong-Ying Lin, Yi-Min Wang et al.· Journal of Chemical Informat...· 0 citations
T cell receptor (TCR) binding to peptides presented by major histocompatibility complex (MHC) molecules is a key step in T cell activation, and forms the basis of adaptive immunity. Predicting this specificity is therefore essential to developing effective TCR-based immunotherapies and vaccines. Despite its clinical relevance, predicting TCR-pMHC specificity for previously unseen peptides remains an open problem, with structural modeling so far the only strategy showing any predictive power in this setting. In this study, we find that this limited performance is substantially driven by label noise in the data used to train and evaluate these methods, an effect that has so far been largely underexplored. Using an AlphaFold3-based pipeline adapted for TCR-pMHC structural modeling, we achieve state-of-the-art specificity prediction, outperforming AlphaFold2.3-based and sequence based methods, and performing at par with the leading Immrep2025 competition submission. Combining this pipeline with a cluster-based denoising algorithm, we show that removing mislabeled points from a large specificity dataset increased binder ranking accuracy by more than 70% relative to the full dataset. Together, these results highlight label noise as a major factor limiting the performance that any method in this field can achieve, and show that combining structural modeling with label denoising substantially improves TCR-pMHC specificity prediction, making such approaches an attractive complement to current sequence-based approaches for refining TCR target selection.
Pilar Ballesteros-Cuartero, J. Lund, Morten Nielsen· bioRxiv· 1 citation· ⚡1
The effective design of T cell receptor (TCR)-based therapies — whether in a cellular or cell-engager format — is crucially dependent on the identification of TCRs with high specificity for the target MHC:peptide (MHCp) complex. This makes screening candidate TCRs for cross-reactivity towards off-target self-epitopes as well as ranking their affinity for the target (and off-targets) key activities on the path from epitope to therapy. Here, we present a novel, high-throughput assay for characterisation of TCR and TCR mimic molecules in terms of fine specificity (binding motif) and broader cross-reactivity towards unrelated epitopes. The assay can screen hundreds of MHCp complexes in a single run and enables easy access to essential TCR characteristics. In addition, the assay can be extended to include avidity measurements to allow TCR affinity ranking towards multiple targets.
TCR-coated beads were stained with pools of DNA barcode-labelled MHC Dextramer reagents carrying positional scanning peptide libraries or selected off-target peptides. Subsequent sequencing of the DNA barcodes associated with TCR binding revealed a detailed picture of the TCR binding preferences. By including MHCp complexes of known affinity towards the TCR, all MHCp could be ranked according to their TCR binding affinity.
Using our high-throughput assay, we were able to simultaneously characterize five TCR molecules in terms of their fine specificity and cross-reactivity, while also ranking the affinity of each TCR towards the screened MHCp complexes.
Here, we demonstrate a novel assay for the characterization of TCR fine specificity, cross-reactivity, and affinity in a simple, high-throughput format. The assay enables easy assessment of TCR specificity towards MHCp complexes and is part of a larger pipeline of assays, which combines T cell/TCR discovery with TCR validation and characterisation, an essential part of the development process for TCR-based therapeutics.
Eurostars
Technological Innovations in Immunology (TECH)
Thomas H. Blicher, A. Schøller, L. Brix· Journal of Immunology· 0 citations
T cell antigen-specific immunity depends on pairwise interactions between T cell receptors and peptide- MHC, yet isolating the TCR-pMHC pairs that drive productive engagement remains a major obstacle for antigen-specific therapeutics and for decoding TCR specificity. We overcome this by co-encoding TCR and pMHC in a single founder cell, then clonally expanding it inside a semi-permeable capsule so that genetically identical daughter cells engage in trans. T cell activation, rather than binding affinity, is used to sort cells with functional pairs, and a single PCR on the clone’s linked genomic library captures both partners. This platform, LINC-seq, recovered known cognate pairs from pooled libraries at up to 95% accuracy and performed simultaneous, library-on-library deep mutational scanning of both partners. Wild- type clonotypes ranked among the top-enriched sequences in complex mixtures, and the screens resolved co-evolutionary epistasis and cross-reactivity rules inaccessible to one-sided mutagenesis. The approach generalizes to any receptor-ligand pair whose trans-engagement drives a reporter.
L. Liu, Seung Won Shin, Kevin M. Joslin et al.· bioRxiv· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
