Use TLimmuno2 Online
Commercially Available TLimmuno2 No-Code Web Server
TLimmuno2: Predicting MHC Class II Antigen Immunogenicity with Transfer Learning
TLimmuno2 is a transfer-learning model that predicts whether an MHC class II-bound peptide (epitope) can elicit a CD4+ T cell response. It was developed by Guangshuai Wang, Tao Wu, Wei Ning, Kaixuan Diao, Xiaoqin Sun, Jinyu Wang, Chenxu Wu, Jing Chen, Dongliang Xu and Xue-Song Liu, and published in Briefings in Bioinformatics in 2023. The authors describe it as open-source Python packages with command-line and library interfaces.
Predicting which peptides will trigger an immune response is central to vaccine design and cancer immunotherapy, where neoantigens presented on MHC class II molecules help activate CD4+ T cells. Binding affinity to MHC alone is only a partial signal for immunogenicity.
Labeled immunogenicity data are limited compared with the large pool of peptide-MHC binding measurements. TLimmuno2 tackles this by first learning from binding affinity data and then transferring that knowledge to the smaller immunogenicity task.
Attribute | Details |
|---|---|
Task | Predict CD4+ T cell immunogenicity of MHC class II peptides |
Architecture | Transfer-learning LSTM model |
Inputs | Peptide (padded to 21 residues) and MHC pseudosequence (34 residues), BLOSUM62-encoded |
Binding training data | 107,008 peptide-MHC II measurements across 71 human MHC II types |
Immunogenicity data | 3,930 positive and 2,478 negative instances from IEDB, plus random human-proteome negatives |
Code | Open-source on GitHub |
How TLimmuno2 Predicts Neoantigen Immunogenicity
Binding affinity pretraining: A binding affinity LSTM is trained first. Its penultimate layer, a 10-dimensional peptide-MHC embedding, is carried forward.
Dual sequence encoders: Two additional LSTMs encode the peptide and the MHC pseudosequence.
Feature fusion: Their outputs are concatenated with the transferred embedding and passed through dense layers of 400 and 200 neurons with a sigmoid output.
Fine-tuning: The combined model is fine-tuned on immunogenicity labels.
Binding threshold: Binding labels use a 500 nM cutoff, and percentile ranks are computed against 90,000 random human peptides.
Positional Importance
In mutational analysis with 100 simulations, peptide positions P4, P5 and P6 were the most influential for the predicted immunogenicity.
TLimmuno2 Performance on Immunogenicity Benchmarks
Model component | Metric | Value |
|---|---|---|
Binding affinity model (independent test) | AUC | 0.7742 (MixMHC2pred 0.6482; NetMHCIIpan 0.7965) |
Immunogenicity model, 5-fold CV | Average AUC | 0.8727 |
Immunogenicity model, held-out test | AUC | 0.8649 |
The authors report better performance than existing models on independent validation sets, identification of immunogenic neoantigens in cancer immunotherapy data, and an MHC II neoantigen-associated immunoediting signal in TCGA pan-cancer data (missense mutations across 33 cancer types).
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 TLimmuno2 on Tamarind Bio
Neoantigen prioritization: Rank candidate neoepitopes by predicted CD4+ immunogenicity in cancer immunotherapy studies.
Vaccine design: Screen peptide candidates for MHC class II immunogenicity before synthesis.
Immunoediting analysis: Apply the authors' approach to study MHC II neoantigen-associated immunoediting in tumor data.
Wild-type versus mutant comparison: Compare predicted immunogenicity of mutant peptides with their wild-type counterparts.
How to Use TLimmuno2 on Tamarind Bio
Log in: Sign in at tamarind.bio and open the TLimmuno2 tool.
Provide peptides: Enter or upload the peptide sequences to evaluate; the model was built for peptides up to 21 residues.
Specify the MHC II allele: Select the HLA class II allele(s) if exposed; the model uses an MHC pseudosequence.
Run the prediction: Launch the job and wait for completion.
Download results: Download the predicted immunogenicity scores.
Interpret scores: Compare scores across candidates to rank them, rather than relying on a single value.
Use downstream: Shortlist peptides for experimental validation such as T cell assays.
Things to Keep in Mind
Scope: The model covers MHC-peptide interaction only; peptide processing, presentation and TCR-pMHC interactions are not modeled.
Label assumptions: Positives are assumed immunogenic if validated in at least one experiment, and random human-proteome negatives may themselves be immunogenic.
Limited data: Immunogenicity training data are limited and the neoepitope dataset is imbalanced.
Binding model: The binding affinity model was trained on in vitro data and does not outperform other tools on its own.
Missing features: Features such as DAI and peptide expression are not included.
Source: Wang G, Wu T, Ning W, Diao K, Sun X, Wang J, Wu C, Chen J, Xu D, Liu X-S. TLimmuno2: predicting MHC class II antigen immunogenicity through transfer learning. Briefings in Bioinformatics 24(3):bbad116, 2023. DOI: 10.1093/bib/bbad116. Code: https://github.com/XSLiuLab/TLimmuno2