PERSPECTIVE article

Front. Immunol., 29 August 2025

Sec. Cancer Immunity and Immunotherapy

Volume 16 - 2025 | https://doi.org/10.3389/fimmu.2025.1651533

AI/ML-empowered approaches for predicting T Cell-mediated immunity and beyond

  • CC

    Cheng-chi Chao 1,2

  • YC

    Yulun Chiu 3,4

  • LY

    Lucas Yeung 1

  • CY

    Cassian Yee 3,4

  • CJ

    Chongming Jiang 1*

  • XS

    Xiling Shen 1,5*

  • 1. Terasaki Institute for Biomedical Innovation, Los Angeles, CA, United States

  • 2. ImmuX Consulting, San Jose, CA, United States

  • 3. Department of Melanoma Medical Oncology, Division of Cancer Medicine, The University of Texas MD Anderson Cancer Center, Houston, TX, United States

  • 4. Department of Immunology, The University of Texas MD Anderson Cancer Center, Houston, TX, United States

  • 5. Department of GI Medical Oncology, The University of Texas MD Anderson Cancer Center, Houston, TX, United States

Abstract

T cells play a dual role in various physiopathological states, capable of eliminating tumors and infected cells, while also playing a pathogenic role when activated by autoantigens, causing self-tissue damage. The regulation of T cell-peptide/major histocompatibility complex (TCR-pMHC) recognition is crucial for maintaining disease balance and treating cancer, infections, and autoimmune diseases. Despite efforts, predictive models of TCR-pMHC specificity are still in the early stages. Inspired by advances in protein structure prediction via deep neural networks, we evaluated AlphaFold 3 (AF3)-based AI computation as a method to predict TCR epitope specificity. We demonstrate that AlphaFold can model TCR-pMHC interactions, distinguishing valid epitopes from invalid ones with increasing accuracy. Immunogenic epitopes can be identified for vaccine development through in silico high-throughput processes. Additionally, higher-affinity and specific T cells can be designed to enhance therapy efficacy and safety. An accurate TCR-pMHC prediction model is expected to greatly benefit T-cell-mediated immunotherapy and aid drug design. Overall, precise prediction of T-cell immunogenicity holds significant therapeutic potential, allowing the identification of peptide epitopes linked to tumors, infections, and autoimmune diseases. Although there is much work to be done before these predictions achieve widespread practical use, we are optimistic that deep learning-based structural modeling is a promising pathway for the generalizable prediction of TCR-pMHC interactions.

In-silico prediction of peptide-MHC binding and TCR recognition

There are many previous interests and attempts about predicting the antigen specificity in the T cell activity (). Models for standalone prediction of peptide and MHC binding have existed for decades, such as NetMHC, NetMHCpan, MHCflurry, IEDB-AR, SYFPEITHI, TEPITOPE, MixMHCpred, DeepHLApan, PickPocket, MARIA, SMM, ARBO-MHC, HBond-MHC, MHCnuggets, PepCNN, BigMHC (), and so on () (, ). Yet, for an assessment of antigen specificity and immunogenicity, the precise interaction of a given TCR to its corresponding pMHC complex should be further considered (). Although tools, such as NetTCR (), IMRex (), ERGO (), TEINet (), AEPCAM (), PanPep (), pMTnet (), TEIM-Res (), PISTE (), BERTrand (), BigMHC (), and HLAIImaster (, ), have been available, few models among them are able to accurately predict the recognition of pMHC complexes by T-cell receptors (TCRs) (). The most precise method to dissect TCR-pMHC interactions involve experimentally generating X-ray crystallography structures, which is a time-consuming and technically demanding process. The primary hurdle in accurately predicting T-cell recognition of pMHC complexes lies in the inherent difficulties of protein structure prediction for TCRs and pMHC complexes. While numerous computational models employing various mechanisms have been developed to predict the structure of proteins like antibodies and TCRs, few of them have achieved satisfactory results (). With the advent of the AI/ML era and models, many algorithms, such as AlphaFold 2 and AlphaFold 3 (43, 44), RoseTTAFold/RFdiffusion (45, 46), TrRosetta (47), HADDOCK (48), DeepFRI (49), CANDOCK (50), and Boltz-2 (51), have brought major improvements to predict protein structures and interactions. The accuracy of computational modeling has greatly improved. We have leveraged these advancements to explore AI/ML-driven computational protein structure prediction. Using our current model system, we can exploratively predict T-cell binding to pMHC complexes with significantly higher docking precision, enhancing reliability and biological relevance (, , , 4346, 5258).

AI/ML-powered computational design for predicting TCR-pMHC recognition

Computationally predicting TCR-pMHC interactions using AI/ML approaches offers a rapid, accurate, and scalable alternative to traditional experimental methods, thereby significantly facilitating antigen discovery and immune response modeling. AF3 is a publicly released model developed by DeepMind, trained extensively on more than 120 million protein sequences from the UniProt database and more than 2.2 million experimentally determined protein structures from the Protein Data Bank (PDB). We utilized the model as-is, without retraining or further fine-tuning. The default hyperparameters provided by AlphaFold 3 were employed, including three cycles of recycling, a multiple sequence alignment (MSA) depth of 256, and a template dropout rate of 15% (43, 44). A comparative analysis of AF3 with other structure prediction tools confirms that AF3 predictions outperform other tools in terms of structural accuracy and reliability (59). The results present a comparative analysis of TCR-pMHC recognition between the experimentally determined x-ray crystallography structure and AF3 computational predictions (Figure 1). The experimentally resolved crystal structure of the TCR-pMHC complex is shown in Figure 1A, serving as a reference for evaluating AF3 prediction accuracy. The crystal structure offers detailed insights into the spatial arrangement and interactions among the T-cell receptor (TCR) and peptide-MHC molecules. AF3’s prediction of TCR binding in the presence of peptide-MHC complex is shown in Figure 1B. This prediction closely mirrors the crystal structure in Figure 1A, demonstrating high accuracy in modeling the ternary complex. This highlights AF3’s ability to effectively predict TCR-pMHC interactions once the peptide is bound to the MHC groove. Conversely, AF3’s prediction of TCR binding to MHC in the absence of the same peptide (SLLMWITQC) is illustrated in Figure 1C. This prediction does not align well with the expected TCR binding conformation shown in Figure 1B. The reduced predictive performance from the AF3-based model highlights the importance of peptide presence for accurate binding predictions. This suggests that the conformation of the peptide-MHC complex is essential for accurate TCR interaction. The presence of the peptide resulted in a higher predicted interface template modeling (ipTM) score compared to its absence in AF3’s TCR-pMHC binding prediction (ipTM = 0.92 vs. 0.54, respectively) in the advanced protein structure prediction analysis, as shown in Figures 1B, D in contrast to Figures 1C, E. These high TM-score values confirm strong agreement between AF3 predictions and the crystal structure for the TCR-pMHC complex. The results demonstrated that the AF3-enabled approach reliably predicts TCR-pMHC interactions, supported by a high correlation with crystal structures and favorable TM-scores (Figures 1B, D). Instead, predictive accuracy decreases notably without peptides (Figures 1C, E). A comparative analysis in more TCR-pMHC binding structures assesses how peptide presence influences TCR-pMHC binding, as reflected in the ipTM values. In this analysis, the same set of 9 TCR-pMHC complexes (6067) were included under both conditions: one group is the TCR-pMHC binding structure with peptides (+Peptides) and the other without peptides (-Peptides). The results demonstrate that the ipTM scores of TCR-pMHC binding structures with peptides are significantly higher than those without peptides (two-sided Wilcoxon tests, p-value =6e-04), as shown in Figure 1F. These findings highlight the significance of achieving accurate TCR-pMHC binding predictions with AF3.

Figure 1

Future perspectives and challenges

AI-driven T-cell-pMHC modeling holds significant potential for drug discovery and clinical applications. However, a major hurdle in the existing models is their inability to accurately predict T-cell recognition of its cognate antigen (6870). Developing in-silico tools to assess T-cell immunogenicity is crucial from both biological and therapeutic standpoints. Computational identification and screening of TCR specificity can greatly advance T-cell-related therapeutics, such as T cell therapy and vaccines, enhancing efficacy and safety (, , 55, 68, 69, 71). A generalizable model of TCR-pMHC interactions can significantly accelerate the identification of dominant antigenic epitopes with high affinity for their respective MHC. Improving predictions of TCR binding to the pMHC complex can help fine-tune TCR affinity and address a key challenge in the field. Accurate predictions of the T-cell-pMHC complex structure can aid in designing agonistic or antagonistic peptide analogs to stimulate tumor-specific or tolerize (auto)antigen-specific T cells (, 48, 7275). For vaccine design, an AI-enabled model can assist in selecting suitable epitopes with strong immunogenicity, helping to accurately identify and validate those capable of activating T cells (76, 77). Moreover, accurate predictions of peptide-MHC interaction are expected to enable more effective assessment of the risk of anti-drug responses in patients, although current models often overestimate the results (, 7274). Ultimately, we hope that such a model will be able to predict T-cell functional activity in an exploratory manner, thereby contributing to the development of potential therapeutic strategies (Table 1).

Table 1

I. Predict the binding of peptide epitopes to their respective MHC
Screen and identify antigenic epitopes with potential strong immunogenicity for vaccine discovery.
Tailor the interaction of peptide-MHC interaction for designated therapeutic outcomes.
II. Envision the interaction of antigen-specific T cells with their corresponding pMHC
Predict TCR specificity from disease-associated T cells for target identification in T-cell therapy.
Improve therapeutic design by refining the T-cell affinity to enhance efficacy of therapeutics.
Reduce unintended cross-reactivity/off-target effects to minimize off-target effects for drug safety assessment.
III. Predict the Magnitude and Nature of T Cell Responses
Assess potential T cell responses to therapeutics, including both immunogenic activation and inhibitory effects.
Evaluate patient risk in developing anti-drug immune responses.
Assess the neoantigen quality of tumors and the associated resistance to immune checkpoint blockade therapy.

Exploring three key facets of therapeutic applications for accurate T cell–pMHC recognition prediction.

While the pharmaceutical and biotech industries have leveraged AI across various stages of drug discovery, primarily focusing on small molecules and antibody drugs, the application of AI/ML and digital biology to T-cell therapeutics remains relatively limited (68, 69, 76). As demonstrated by the results above, our advanced protein prediction modeling approach enables accurate prediction of T-cell-pMHC interactions, offering significant potential to enhance T-cell-mediated therapies. Despite advancements, AI-assisted protein design and protein-protein interactions (PPI) still face significant unresolved challenges (Table 2). Tools like AF3, Rosetta, and Boltz-2 have revolutionized protein engineering, yet several critical obstacles remain, as outlined below. One major challenge is the limited availability of high-quality data, especially for underrepresented antigens, rare HLA alleles, and paired TCR alpha and beta chains (, 7880). The lower the quality and quantity of training data available to AI systems, the less reliable their predictions of binding interactions become. Additionally, TCRs naturally exhibit a wide range of binding affinities and can be polyspecific, making it difficult to train models that accurately capture their complex interaction profiles (43, 44, 46, 52, 58, 81). A further complication arises from protein conformational dynamics. Proteins exist in multiple conformations; they open, close, twist, and bend. These conformational changes depend on factors such as temperature, pH, chemical environment, and interactions with other molecules (, , 55). Moreover, TCR binding affinity alone is not sufficient to guarantee a functional immune response. A robust response requires a complex interplay of factors, including antigen processing and presentation, TCR binding,as well as T cell activation, differentiation and the diseased microenvironment (8285). While AlphaFold can help discriminate between correct and incorrect binding partners, it doesn’t directly predict binding affinity in a quantitative way. Additional modeling approaches, such as Rosetta, are required to calculate binding energy changes, which can then be used to predict the effects of mutations on TCR affinity and correlate them with binding affinity. Ultimately, the development of a reliable and high-throughput screening system is essential for identifying effective therapeutic candidates in drug discovery. While our exploratory model system is an initial step, we aspire for it to drive future advancements in T-cell-based therapeutics.

Table 2

I. Limited availability of diverse, high-quality data for training
Only a small fraction of potential TCR-ligand pairs available overall for model training.
Data on a diverse array of epitopes binding to TCRs needs to be generated. Most antigens reported to bind TCRs are viral, comprising the majority of TCR-antigen pairs.
Current datasets are dominated by antigens presented by common HLA alleles, with few under-represented HLA alleles included.
II. Focusing solely on peptide-MHC binding without considering TCR interactions
A strong peptide-MHC interaction may be necessary for T cell activation, but it is not sufficient to guarantee an immune response.
III. Focusing solely on the β-chain CDR3 loops of TCR sequence information
Both α and β chains contribute to antigen recognition and specificity. Incorporating both chains improves predictive performance, but chain pairing information is largely missing from current datasets.
IV. Lack of INFO about the polyspecificity of individual TCRs
TCRs can exhibit both specificity and promiscuity. Models that assume a given TCR recognizes only a single cognate epitope oversimplify the complexity of TCR-antigen interactions.
V. Stronger TCR binding affinity alone does not necessarily translate to a stronger functional immune response
Robust predictions of TCR specificity require a complex interplay of factors, including antigen processing and presentation, TCR binding, T cell activation, differentiation, and the diseased microenvironment.
VI. The influence of thymic selection and self-peptide presentation is overlooked
The naive immune repertoire formation is an under-explored area in TCR specificity prediction. The high affinity or immunogenicity of TCR predicted may not exist in periphery due to thymic deletion.
VII. Limitations in Direct Quantification of Binding Affinity Prediction
While AlphaFold can help discriminate between correct and incorrect binding partners, it doesn’t directly predict binding affinity in a quantitative way.
VIII. The effect of hallucination of the AL/ML models will impact the precision of Binding Affinity Prediction
The key factors, such as data quality, model regularization, fine-tuning, and better supervision, contribute to hallucination in the AL/ML models during binding affinity prediction.

The major challenges in predicting TCR immunogenicity in silico.

Statements

Data availability statement

The original contributions presented in the study are included in the article/supplementary material. Further inquiries can be directed to the corresponding authors.

Author contributions

C-cC: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Writing – original draft, Writing – review & editing. YC: Data curation, Formal analysis, Validation, Writing – review & editing. LY: Writing – review & editing, Data curation, Validation, Investigation, Visualization. CY: Conceptualization, Data curation, Resources, Validation, Writing – review & editing. CJ: Data curation, Formal analysis, Investigation, Methodology, Software, Supervision, Validation, Visualization, Writing – original draft, Writing – review & editing. XS: Funding acquisition, Resources, Supervision, Writing – original draft, Writing – review & editing.

Funding

The authors declare financial support was received for the research and/or publication of this article. This work is supported by the National Institutes of Health, United States (NIH) R01 DK119795, R35 GM122465, and the Cancer Prevention Research Institute of Texas (CPRIT) (RR240007).

Acknowledgments

This work is supported by the National Institutes of Health, United States (NIH) R01 DK119795, R35 GM122465, and the Cancer Prevention Research Institute of Texas (CPRIT) (RR240007). Thank Xiuying Li for useful suggestions.

Conflict of interest

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Generative AI statement

The authors declare that no Generative AI was used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

References

  • 1

    GlanvilleJHuangHNauAHattonOWagarLERubeltFet al. Identifying specificity groups in the T cell receptor repertoire. Nat Publishing Group. (2017) 547:94–8. doi: 10.1038/nature22976

  • 2

    GeorgeJTKesslerDALevineH. Effects of thymic selection on T cell recognition of foreign and tumor antigenic peptides. Proc Natl Acad Sci. (2017) 114:E7875–81. doi: 10.1073/pnas.1708573114

  • 3

    CalisJMaybenoMGreenbaumJ a.WeiskopfDDe SilvaADSetteAet al. Properties of MHC class I presented peptides that enhance immunogenicity. PloS Comput Biol. (2013) 9:e1003266. doi: 10.1371/journal.pcbi.1003266

  • 4

    DashPFiore-gartlandAJHertzTWangGCSharmaSSouquetteAet al. Quantifiable predictive features define epitope- specific T cell receptor repertoires. Nat Publishing Group. (2017) 547:8993. doi: 10.1038/nature22383

  • 5

    ZhaoWSherX. Systematically benchmarking peptide-MHC binding predictors: From synthetic to naturally processed epitopes. PloS Comput Biol. (2018) 14:128. doi: 10.1371/journal.pcbi.1006457

  • 6

    ParkhurstMGoffSLLoweryFJBeyerRKHalasHRobbinsPFet al. Adoptive transfer of personalized neoantigen-reactive TCR-transduced T cells in metastatic colorectal cancer: phase 2 trial interim results. Nat Med. (2024) 30:2586–95. doi: 10.1038/s41591-024-03109-0

  • 7

    FengZChenJHaiYPangXZhengKXieCet al. Sliding-attention transformer neural architecture for predicting T cell receptor–antigen–human leucocyte antigen binding. Nat Mach Intell. (2024) 6:1921. doi: 10.1038/s42256-024-00901-y

  • 8

    KarnaukhovVKShcherbininDSChugunovAOChudakovDMEfremovRGZvyaginIV. Structure-based prediction of T cell receptor recognition of unseen epitopes using TCRen. Nat Comput Sci. (2024) 4:510–21. doi: 10.1038/s43588-024-00653-0

  • 9

    SaethangTHiroseOKimkongITranVADangXTNguyenLATet al. PAAQD: Predicting immunogenicity of MHC class I binding peptides using amino acid pairwise contact potentials and quantum topological molecular similarity descriptors. J Immunol Methods. (2013) 387:293302. doi: 10.1016/j.jim.2012.09.016

  • 10

    TickotskyNSagivTPriluskyJShifrutEFriedmanN. McPAS-TCR: A manually curated catalogue of pathology-associated T cell receptor sequences. Bioinformatics. (2017) 33:2924–9. doi: 10.1093/bioinformatics/btx286

  • 11

    RasmussenMFenoyEHarndahlMKristensenABNielsenIKNielsenMet al. Pan-specific prediction of peptide-MHC class I complex stability, a correlate of T cell immunogenicity. J Immunol. (2016) 197:1517–24. doi: 10.4049/jimmunol.1600582

  • 12

    BirnbaumMEMendozaJLSethiDKDongSGlanvilleJDobbinsJet al. Deconstructing the peptide-MHC specificity of T cell recognition. Cell. (2014) 157(5):1073–87. doi: 10.1016/j.cell.2014.03.047

  • 13

    HuangJZengXSigalNLundPJSuLFHuangHet al. Detection, phenotyping, and quantification of antigen-specific T cells using a peptide-MHC dodecamer. (2016) 113(13):E1890–7. doi: 10.1073/pnas.1602488113

  • 14

    MikhaylovVBrambleyCAKellerGLJArbuisoAGWeissLIBakerBMet al. Accurate modeling of peptide-MHC structures with AlphaFold. Structure. (2024) 32:228241.e4. doi: 10.1016/j.str.2023.11.011

  • 15

    WuJWangWZhangJZhouBZhaoWSuZet al. DeepHLApan: A deep learning approach for neoantigen prediction considering both HLA-peptide binding and immunogenicity. Front Immunol. (2019) 10:2559. doi: 10.3389/fimmu.2019.02559

  • 16

    HuppaJBAxmannMMoMADavisMMBrameshuberMKleinLOet al. TCR–peptide–MHC interactions in situ show accelerated kinetics and increased affinity. Nature (2010) 463:963–7. doi: 10.1038/nature08746

  • 17

    ReinherzELTanKTangLKernPLiuJXiongYet al. The crystal structure of a T cell receptor in complex with peptide and MHC class II. Science. (1999) 286:1913–21. doi: 10.1126/science.286.5446.1913

  • 18

    GaoYGaoYFanYZhuCWeiZZhouCet al. Pan-Peptide Meta Learning for T-cell receptor–antigen binding recognition. Nat Mach Intell. (2023) 5:236–49. doi: 10.1038/s42256-023-00619-3

  • 19

    SpringerIBesserHTickotsky-MoskovitzNDvorkinSLouzounY. Prediction of specific TCR-peptide binding from large dictionaries of TCR-peptide pairs. Front Immunol. (2020) 11:1803. doi: 10.3389/fimmu.2020.01803

  • 20

    AlbertBAYangYShaoXMSinghDSmitKNAnagnostouVet al. Deep neural networks predict class I major histocompatibility complex epitope presentation and transfer learn neoepitope immunogenicity. Nat Mach Intell. (2023) 5:861–72. doi: 10.1038/s42256-023-00694-6

  • 21

    VitaRMahajanSOvertonJADhandaSKMartiniSCantrellJRet al. The immune epitope database (IEDB): 2018 update. Nucleic Acids Res. (2019) 47:D339–43. doi: 10.1093/nar/gky1006

  • 22

    O’DonnellTJRubinsteynABonsackMRiemerABLasersonUHammerbacherJ. MHCflurry: open-source class I MHC binding affinity prediction. Cell Syst. (2018) 7:129132.e4. doi: 10.1016/j.cels.2018.05.014

  • 23

    RammenseeHBachmannJEmmerichNPBachorOAStevanovićS. SYFPEITHI: database for MHC ligands and peptide motifs. Immunogenetics. (1999) 50:213–9. doi: 10.1007/s002510050595

  • 24

    WuJLiJChenSZhouZ. DeepHLApan: A deep learning approach for the prediction of peptide-HLA binding and immunogenicity. Methods Mol Biol. (2024) 2809:237–44. doi: 10.1007/978-1-0716-3874-3_15

  • 25

    JensenKKAndreattaMMarcatiliPBuusSGreenbaumJAYanZet al. Improved methods for predicting peptide binding affinity to MHC class II molecules. Immunology. (2018) 154:394406. doi: 10.1111/imm.12889

  • 26

    DönnesPElofssonA. Prediction of MHC class I binding peptides, using SVMHC. BMC Bioinf. (2002) 3:25. doi: 10.1186/1471-2105-3-25

  • 27

    ZhangHLundONielsenM. The PickPocket method for predicting binding specificities for receptors based on receptor pocket similarities: application to MHC-peptide binding. Bioinformatics. (2009) 25:1293–9. doi: 10.1093/bioinformatics/btp137

  • 28

    WieczorekMAbualrousETStichtJÁlvaro-BenitoMStolzenbergSNoéFet al. Major histocompatibility complex (MHC) class I and MHC class II proteins: conformational plasticity in antigen presentation. Front Immunol. (2017) 8:292. doi: 10.3389/fimmu.2017.00292

  • 29

    ChandraASharmaADehzangiITsunodaTSattarA. PepCNN deep learning tool for predicting peptide binding residues in proteins using sequence, structural, and language model features. Sci Rep. (2023) 13:20882. doi: 10.1038/s41598-023-47624-5

  • 30

    SalesAPTomarasGDKeplerTB. Improving peptide-MHC class I binding prediction for unbalanced datasets. BMC Bioinf. (2008) 9:385. doi: 10.1186/1471-2105-9-385

  • 31

    HudsonDFernandesRABashamMOggGKoohyH. Can we predict T cell specificity with digital biology and machine learning? Nat Rev Immunol. (2023) 23:511–21. doi: 10.1038/s41577-023-00835-3

  • 32

    CroceGBobisseSMorenoDLSchmidtJGuillamePHarariAet al. Deep learning predictions of TCR-epitope interactions reveal epitope-specific chains in dual alpha T cells. Nat Commun. (2024) 15:3211. doi: 10.1038/s41467-024-47461-8

  • 33

    KohlgruberACDezfulianMHSieBMWangCIKulaTLasersonUet al. High-throughput discovery of MHC class I- and II-restricted T cell epitopes using synthetic cellular circuits. Nat Biotechnol. (2024) 43:623–34. doi: 10.1038/s41587-024-02248-6

  • 34

    MontemurroASchusterVPovlsenHRBentzenAKJurtzVChronisterWDet al. NetTCR-2.0 enables accurate prediction of TCR-peptide binding by using paired TCRα and β sequence data. Commun Biol. (2021) 4:113. doi: 10.1038/s42003-021-02610-3

  • 35

    MorisPDe PauwJPostovskayaAGielisSDe NeuterNBittremieuxWet al. Current challenges for unseen-epitope TCR interaction prediction and a new perspective derived from image classification. Brief Bioinform. (2021) 22:112. doi: 10.1093/bib/bbaa318

  • 36

    JiangYHuoMCheng LiS. TEINet: a deep learning framework for prediction of TCR-epitope binding specificity. Brief Bioinform. (2023) 24:bbad086. doi: 10.1093/bib/bbad086

  • 37

    ChenJZhaoBLinSSunHMaoXWangMet al. TEPCAM: Prediction of T-cell receptor-epitope binding specificity via interpretable deep learning. Protein Sci. (2024) 33:e4841. doi: 10.1002/pro.4841

  • 38

    LuTZhangZZhuJWangYJiangPXiaoXet al. Deep learning-based prediction of the T cell receptor-antigen binding specificity. Nat Mach Intell. (2021) 3:864–75. doi: 10.1038/s42256-021-00383-2

  • 39

    PengXLeiYFengPJiaLMaJZhaoDet al. Characterizing the interaction conformation between T-cell receptors and epitopes with deep learning. Nat Mach Intell. (2023) 5:395407. doi: 10.1038/s42256-023-00634-4

  • 40

    MyronovAMazzoccoGKrólPPlewczynskiD. BERTrand-peptide:TCR binding prediction using Bidirectional Encoder Representations from Transformers augmented with random TCR pairing. Bioinformatics. (2023) 39:19. doi: 10.1093/bioinformatics/btad468

  • 41

    YangQXuLDongWLiXWangKDongSet al. HLAIImaster: a deep learning method with adaptive domain knowledge predicts HLA II neoepitope immunogenic responses. Brief Bioinform. (2024) 25:bbae302. doi: 10.1093/bib/bbae302

  • 42

    XuLYangQDongWLiXWangKDongSet al. Meta learning for mutant HLA class I epitope immunogenicity prediction to accelerate cancer clinical immunotherapy. Brief Bioinform. (2024) 26:bbae625. doi: 10.1093/bib/bbae625

  • 43

    JumperJEvansRPritzelAGreenTFigurnovMRonnebergerOet al. Highly accurate protein structure prediction with AlphaFold. Nature. (2021) 596:583–89. doi: 10.1038/s41586-021-03819-2

  • 44

    GuanXTangQRenWChenMWangWWolynesPGet al. Predicting protein conformational motions using energetic frustration analysis and AlphaFold2. Proc Natl Acad Sci U S A. (2024) 121(35):111. doi: 10.1073/pnas.2410662121/-/DCSupplemental.Published

  • 45

    Vázquez TorresSLeungPJYVenkateshPLutzIDHinkFHuynhHHet al. De novo design of high-affinity binders of bioactive helical peptides. Nature. (2024) 626:435–42. doi: 10.1038/s41586-023-06953-1

  • 46

    WatsonJLJuergensDBennettNRTrippeBLYimJEisenachHEet al. De novo design of protein structure and function with RFdiffusion. Nature. (2023) 620:1089–100. doi: 10.1038/s41586-023-06415-8

  • 47

    AdelborgKSzentkútiPHenriksenJEThomsenRWPedersenLSundbøllJet al. Cohort profile: the Funen Diabetes Database-a population-based cohort of patients with diabetes in Denmark. BMJ Open. (2020) 10:e035492. doi: 10.1136/bmjopen-2019-035492

  • 48

    PeacockTChainB. Information-driven docking for TCR-pMHC complex prediction. Front Immunol. (2021) 12:686127. doi: 10.3389/fimmu.2021.686127

  • 49

    PagiatakisCMusolinoEGornatiRBernardiniGPapaitR. Epigenetics of aging and disease: a brief overview. Aging Clin Exp Res. (2021) 33:737–45. doi: 10.1007/s40520-019-01430-0

  • 50

    FeghaliJKimJGamiARapaportSCaplanJMMcDougallCGet al. Monocyte-based inflammatory indices predict outcomes following aneurysmal subarachnoid hemorrhage. Neurosurg Rev. (2021) 44:3499–507. doi: 10.1007/s10143-021-01525-1

  • 51

    PassaroSCorsoGWohlwendJReveizMThalerSSomnathVRet al. Boltz-2: Towards Accurate and Efficient Binding Affinity Prediction. (2025) 2025.06.14.659707. doi: 10.1101/2025.06.14.659707

  • 52

    McMasterBThorpeCOggGDeaneCMKoohyH. Can AlphaFold’s breakthrough in protein structure help decode the fundamental principles of adaptive cellular immunity? Nat Methods. (2024) 21:766–76. doi: 10.1038/s41592-024-02240-7

  • 53

    BradleyP. Structure-based prediction of T cell receptor:peptide-MHC interactions. Elife. (2023) 12:e82813. doi: 10.7554/eLife.82813

  • 54

    ChengJNovatiGPanJBycroftCŽemgulyteAApplebaumTet al. Accurate proteome-wide missense variant effect prediction with AlphaMissense. Sci (1979). (2023) 381:eadg7492. doi: 10.1126/science.adg7492

  • 55

    MarzellaDFPariziFMvan TilborgDRenaudNSybrandiDBuzatuRet al. PANDORA: A fast, anchor-restrained modelling protocol for peptide: MHC complexes. Front Immunol. (2022) 13:878762. doi: 10.3389/fimmu.2022.878762

  • 56

    QiaoZNieWVahdatAMillerTFAnandkumarA. State-specific protein–ligand complex structure prediction with a multiscale deep generative model. Nat Mach Intell. (2024) 6:195208. doi: 10.1038/s42256-024-00792-z

  • 57

    AgarwalVMcShanAC. The power and pitfalls of AlphaFold2 for structure prediction beyond rigid globular proteins. Nat Chem Biol. (2024) 20:950–9. doi: 10.1038/s41589-024-01638-w

  • 58

    WuDYinRChenGRibeiro-FilhoHVCheungMRobbinsPFet al. Structural characterization and AlphaFold modeling of human T cell receptor recognition of NRAS cancer neoantigens. bioRxiv. (2024) 2024.05.21.595215:2024.05.21.595215. doi: 10.1126/sciadv.adq6150

  • 59

    XuSFengQQiaoLWuHShenTChengYet al. FoldBench: An All-atom Benchmark for Biomolecular Structure Prediction. (2025) 2025.05.22.655600. doi: 10.1101/2025.05.22.655600

  • 60

    SamiMRizkallahPJDunnSMolloyPMoyseyRVuidepotAet al. Crystal structures of high affinity human T-cell receptors bound to peptide major histocompatibility complex reveal native diagonal binding geometry. Protein Eng Des Sel. (2007) 20:397403. doi: 10.1093/protein/gzm033

  • 61

    YangXGaoMChenGPierceBGLuJWengN-Pet al. Structural basis for clonal diversity of the public T cell response to a dominant human cytomegalovirus epitope. J Biol Chem. (2015) 290:29106–19. doi: 10.1074/jbc.M115.691311

  • 62

    SongIGilAMishraRGhersiDSelinLKSternLJ. Broad TCR repertoire and diverse structural solutions for recognition of an immunodominant CD8+ T cell epitope. Nat Struct Mol Biol. (2017) 24:395406. doi: 10.1038/nsmb.3383

  • 63

    IshizukaJStewart-JonesGBEvan der MerweABellJIMcMichaelAJJonesEY. The structural dynamics and energetics of an immunodominant T cell receptor are programmed by its Vbeta domain. Immunity. (2008) 28:171–82. doi: 10.1016/j.immuni.2007.12.018

  • 64

    DingYHSmithKJGarbocziDNUtzUBiddisonWEWileyDC. Two human T cell receptors bind in a similar diagonal mode to the HLA-A2/Tax peptide complex using different TCR amino acids. Immunity. (1998) 8:403–11. doi: 10.1016/s1074-7613(00)80546-4

  • 65

    BulekAMColeDKSkoweraADoltonGGrasSMaduraFet al. Structural basis for the killing of human beta cells by CD8(+) T cells in type 1 diabetes. Nat Immunol. (2012) 13:283–9. doi: 10.1038/ni.2206

  • 66

    MaduraFRizkallahPJHollandCJFullerABulekAGodkinAJet al. Structural basis for ineffective T-cell responses to MHC anchor residue-improved “heteroclitic” peptides. Eur J Immunol. (2015) 45:584–91. doi: 10.1002/eji.201445114

  • 67

    WooldridgeLEkeruche-MakindeJvan den BergHASkoweraAMilesJJTanMPet al. A single autoimmune T cell receptor recognizes more than a million different peptides. J Biol Chem. (2012) 287:1168–77. doi: 10.1074/jbc.M111.289488

  • 68

    PunFWOzerovIVZhavoronkovA. AI-powered therapeutic target discovery. Trends Pharmacol Sci. (2023) 44:561–72. doi: 10.1016/j.tips.2023.06.010

  • 69

    QureshiRIrfanMGondalTMKhanSWuJHadiMUet al. AI in drug discovery and its clinical relevance. Heliyon. (2023) 9:e17575. doi: 10.1016/j.heliyon.2023.e17575

  • 70

    RollinsZACurtisMBGeorgeSCFallerR. A computational strategy for the rapid identification and ranking of patient-specific T cell receptors bound to neoantigens. Macromol Rapid Commun. (2024) 45:e2400225. doi: 10.1002/marc.202400225

  • 71

    PhamMNNguyenT-NTranLSNguyenQ-TBNguyenT-PHPhamTMQet al. epiTCR: a highly sensitive predictor for TCR-peptide binding. Bioinformatics. (2023) 39:btad284. doi: 10.1093/bioinformatics/btad284

  • 72

    DengLLyCAbdollahiSZhaoYPrinzIBonnS. Performance comparison of TCR-pMHC prediction tools reveals a strong data dependency. Front Immunol. (2023) 14:1128326. doi: 10.3389/fimmu.2023.1128326

  • 73

    KimSHLeeBRKimS-MKimSKimM-SKimJet al. The identification of effective tumor-suppressing neoantigens using a tumor-reactive TIL TCR-pMHC ternary complex. Exp Mol Med. (2024) 56:1461–71. doi: 10.1038/s12276-024-01259-2

  • 74

    YadavSVoraDSSundarDDhanjalJK. TCR-ESM: Employing protein language embeddings to predict TCR-peptide-MHC binding. Comput Struct Biotechnol J. (2024) 23:165–73. doi: 10.1016/j.csbj.2023.11.037

  • 75

    ZhangJMaWYaoH. Accurate TCR-pMHC interaction prediction using a BERT-based transfer learning method. Brief Bioinform. (2023) 25:bbad436. doi: 10.1093/bib/bbad436

  • 76

    KumarADixitSSrinivasanKMDVincentPMDR. Personalized cancer vaccine design using AI-powered technologies. Front Immunol. (2024) 15:1357217. doi: 10.3389/fimmu.2024.1357217

  • 77

    BhattacharyaMAlshammariAAlharbiMDhamaKLeeS-SChakrabortyC. A novel mutation-proof, next-generation vaccine to fight against upcoming SARS-CoV-2 variants and subvariants, designed through AI enabled approaches and tools, along with the machine learning based immune simulation: A vaccine breakthrough. Int J Biol Macromol. (2023) 242:124893. doi: 10.1016/j.ijbiomac.2023.124893

  • 78

    SarkizovaSKlaegerSLePMLiLWOliveiraGKeshishianHet al. A large peptidome dataset improves HLA class I epitope prediction across most of the human population. Nat Biotechnol. (2020) 38:199209. doi: 10.1038/s41587-019-0322-9

  • 79

    JiangCLiJZhangWZhuangZLiuGHongWet al. Potential association factors for developing effective peptide-based cancer vaccines. Front Immunol. (2022) 13:931612. doi: 10.3389/fimmu.2022.931612

  • 80

    KawakitaSShenAChaoC-CWangZChengSLiBet al. An integrated database of experimentally validated major histocompatibility complex epitopes for antigen-specific cancer therapy. Antib Ther. (2024) 7:177–86. doi: 10.1093/abt/tbae011

  • 81

    RileyTPKellerGLJSmithARDavancazeLMArbuisoAGDevlinJRet al. Structure based prediction of neoantigen immunogenicity. Front Immunol. (2019) 10:2047. doi: 10.3389/fimmu.2019.02047

  • 82

    GálvezJGálvezJJGarcía-PeñarrubiaP. Is TCR/pMHC affinity a good estimate of the T-cell response? An answer based on predictions from 12 phenotypic models. Front Immunol. (2019) 10:349. doi: 10.3389/fimmu.2019.00349

  • 83

    al-RamadiBKJelonekMTBoydLFMarguliesDHBothwellAL. Lack of strict correlation of functional sensitization with the apparent affinity of MHC/peptide complexes for the TCR. J Immunol. (1995) 155:662–73. doi: 10.4049/jimmunol.155.2.662

  • 84

    BhattacharyyaNDCounoupasCDanielLZhangGCookSJCootesTAet al. TCR affinity controls the dynamics but not the functional specification of the antimycobacterial CD4+ T cell response. J Immunol. (2021) 206:2875–87. doi: 10.4049/jimmunol.2001271

  • 85

    BräunleinELupoliGFüchslFAbualrousETde Andrade KrätzigNGosmannDet al. Functional analysis of peripheral and intratumoral neoantigen-specific TCRs identified in a patient with melanoma. J Immunother Cancer. (2021) 9:e002754. doi: 10.1136/jitc-2021-002754

Summary

Keywords

TCR-pMHC recognition, AI/ML-driven structure prediction, immunogenicity modeling, T-cell therapy design, protein-protein interactions

Citation

Chao C, Chiu Y, Yeung L, Yee C, Jiang C and Shen X (2025) AI/ML-empowered approaches for predicting T Cell-mediated immunity and beyond. Front. Immunol. 16:1651533. doi: 10.3389/fimmu.2025.1651533

Received

21 June 2025

Accepted

04 August 2025

Published

29 August 2025

Volume

16 - 2025

Edited by

Hyejin Choi, Johnson & Johnson, United States

Reviewed by

Qiang Yang, Harbin Institute of Technology, China

Updates

Copyright

*Correspondence: Xiling Shen, ; Chongming Jiang,

Disclaimer

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.

Outline

Figures

Cite article

Copy to clipboard


Export citation file


Share article

Article metrics