Use TCRModel2 Online
Commercially Available TCRModel2 No-Code Web Server
TCRmodel2: High-Resolution TCR-Peptide-MHC Structure Prediction with Adapted AlphaFold
TCRmodel2 is a web server and method for predicting the 3D structure of T cell receptor (TCR) complexes with peptide-MHC (pMHC) directly from sequence. It was developed by Rui Yin, Helder V. Ribeiro-Filho, Valerie Lin, Ragul Gowthaman, Melyssa Cheung and Brian G. Pierce, with authors affiliated with the University of Maryland Institute for Bioscience and Biotechnology Research, the University of Maryland, College Park, and the Brazilian Biosciences National Laboratory. The paper describes it as an extension of the earlier TCRmodel server, which modeled unbound TCRs.
TCRs recognize peptides presented by MHC molecules, and this recognition underlies T cell immunity. Experimental structures of TCR-pMHC complexes are limited, so reliable computational models are valuable for studying specificity. TCRmodel2 adapts AlphaFold to this problem, with TCR- and MHC-focused sequence searches, TCR-specific templates and tailored confidence scores.
Property | Detail (from the paper) |
|---|---|
Task | Model TCR-peptide-MHC complexes (Class I or II) or unbound TCRs from sequence |
Inputs | TCR, peptide and MHC sequences (direct entry or FASTA), or TCR and MHC genes chosen from human or mouse sets |
Outputs | Five ranked models per job as downloadable PDB files, with confidence scores and an interactive viewer |
Speed | About 15 minutes per job on an NVIDIA Titan RTX GPU |
Publication | Nucleic Acids Research, Web Server issue, 2023 |
How TCRmodel2 Models T Cell Receptor Complexes
Focused sequence search: sequence databases were reduced to TCR- and MHC-related hits to speed up multiple sequence alignment construction.
TCR templates: templates are found using the TCR sequence rather than an alignment, which the authors found avoided matching non-TCR immunoglobulin structures.
pMHC templates: peptide-MHC structures from TCR3d serve as templates: 884 Class I and 44 Class II.
Input handling: TCR variable domains are identified with ANARCI, and Class II peptides are truncated to 11-mers based on NetMHCIIPan 9-mer cores.
Scoring and relaxation: model confidence combines pTM and ipTM (0.2 x pTM + 0.8 x ipTM), alongside TCR-pMHC interface ipTM and CDR3 pLDDT. Amber relaxation is optional.
TCRmodel2 Accuracy, Confidence Scores and Speed
On an initial benchmark of 48 complexes (32 Class I, 16 Class II), TCRmodel2 produced Medium or High CAPRI-accuracy models for over 50% of cases and outperformed AlphaFold 2.2 and ImmuneScape. On an updated 20-complex benchmark (14 Class I, 6 Class II), TCRmodel2 with AlphaFold 2.3 reached 20% High-accuracy models and outperformed AlphaFold 2.3 and TCRDock. For unbound TCRs, accuracy was comparable to AlphaFold and better than TCRmodel, especially for CDR3 loops.
Metric | Reported value |
|---|---|
Confidence separating Medium/High from Incorrect models | AUC = 0.97 |
Correlation of confidence with DockQ | r = 0.75 |
Suggested confidence cutoffs | 0.85 or higher likely accurate; 0.49 or lower likely inaccurate |
Runtime | About 15 minutes versus roughly 5-7 hours for standard AlphaFold multimer runs |
Application to VDJdb complexes
For VDJdb TCR-pMHC pairs, 30% of Class I and 47% of Class II top models had confidence of 0.85 or higher; at 0.75 or higher the figures were 77% and 89%.
What is Tamarind Bio?
Tamarind Bio is a no-code bioinformatics platform built to give life scientists and researchers access to powerful computational tools. Many cutting-edge machine learning models are hard to deploy and use. Tamarind provides an intuitive, web-based environment that removes the complexity of high-performance computing, software dependencies and command-line interfaces.
The platform is designed for biologists, chemists and other researchers who may not have a background in programming or cloud infrastructure but want to run models on their own data. Key features include:
A user-friendly graphical interface for setting up and launching experiments
A robust API for integration into existing research pipelines
An automated system for managing and scaling computational resources
Tamarind treats information and data security as a top priority, as detailed in its Trust Center and Terms of Service.
Accelerating Discovery with TCRmodel2 on Tamarind Bio
Studying TCR specificity: model how a given TCR engages a peptide-MHC to generate structural hypotheses about recognition.
Triaging TCR-pMHC pairs: use confidence scores to prioritize pairs, as the authors did for VDJdb entries.
Unbound TCR modeling: build TCR structures, including CDR3 loops, when no experimental structure exists.
TCR-mimic antibodies: the authors propose applying the approach here, though they note this was not tested.
How to Use TCRmodel2 on Tamarind Bio
Open the tool: log in to tamarind.bio and open the TCRmodel2 tool.
Choose the mode: select TCR-pMHC complex (Class I or II) or unbound TCR, if exposed.
Provide sequences: enter the TCR chain sequences along with the peptide and MHC sequences, as the paper's server accepts.
Set options: choose any exposed options, such as whether to apply Amber relaxation.
Run the job: submit and wait; the paper reports about 15 minutes per complex on its GPU server.
Download models: retrieve the ranked PDB models and their confidence values.
Interpret confidence: favor models at or above 0.85 and treat those at or below 0.49 with caution, per the authors' cutoffs.
Things to Keep in Mind
TCRs that engage CD1 or MR1 are not modeled, because AlphaFold does not support their lipid and small-molecule antigens.
High-accuracy success was limited, at 20% on the updated benchmark.
Confidence cutoffs derive from a 48-case benchmark and the authors note that confidence estimates need improvement.
The method depends on AlphaFold and its training data cutoffs.
Source: Yin, R., Ribeiro-Filho, H. V., Lin, V., Gowthaman, R., Cheung, M., Pierce, B. G. TCRmodel2: high-resolution modeling of T cell receptor recognition using deep learning. Nucleic Acids Research 51(W1), W569-W576 (2023). DOI: 10.1093/nar/gkad356. Web server: https://tcrmodel.ibbr.umd.edu