Use DeepImmuno Online
Commercially Available DeepImmuno No-Code Web Server
DeepImmuno: Deep Learning Prediction and Generation of Immunogenic Peptides for T-Cell Immunity
DeepImmuno is a deep learning framework for the prediction and generation of immunogenic peptides for T-cell immunity. It was developed by Guangyuan Li, Balaji Iyer, V B Surya Prasath, Yizhao Ni and Nathan Salomonis at the University of Cincinnati and Cincinnati Children's Hospital Medical Center, and published in Briefings in Bioinformatics in 2021. It has two parts: DeepImmuno-CNN, a convolutional neural network that scores the immunogenicity of peptide-MHC class I pairs, and DeepImmuno-GAN, a Wasserstein generative adversarial network that proposes new peptide sequences.
Many tools predict whether a peptide binds an MHC molecule, but binding alone does not mean that a peptide triggers a T-cell response. The authors focus on immunogenicity itself, using experimental T-cell assay data, and note that existing approaches, including theirs, have low specificity.
Property | Detail |
|---|---|
Task | Immunogenicity prediction for peptide-HLA class I pairs; generation of immunogenic-like peptides |
Inputs | Peptide sequence (9-mers and 10-mers) and HLA allele |
Output | Continuous immunogenicity score from 0 to 1 |
Training data | 8,971 cleaned T-cell assay instances from IEDB (4,059 positive, 4,912 negative) |
Venue | Briefings in Bioinformatics 22(6), bbab160, 2021 |
How DeepImmuno Predicts Peptide Immunogenicity
Peptide and HLA encoding: peptides are encoded with a reduced PCA space of 566 AAindex1 physicochemical features, together with HLA paratope information.
CNN scoring: two consecutive convolutional layers followed by two fully connected layers produce the immunogenicity score.
Evidence-weighted training: a beta-binomial model weights each training example by the strength of its experimental evidence. Scores of 0.5 or higher were treated as immunogenic in the comparisons.
Peptide generation: DeepImmuno-GAN turns random noise into pseudo-peptide sequences that mimic immunogenic peptides.
DeepImmuno Benchmark Results
Test | Result |
|---|---|
10-fold cross-validation on IEDB | Average auROC 0.85 and auPR 0.81 |
Tumor neoantigens (608 tested) | 29 of 35 immunogenic neoantigens identified (83%), versus 63% for IEDB and 34% for DeepHLApan |
SARS-CoV-2 sensitivity | 68% (convalescent) and 88% (unexposed), versus 52% and 38% for IEDB |
GAN-generated peptides | About 40% of random sequences scored immunogenic, rising to 67% after training |
Which peptide positions matter
Occlusion analysis showed that positions P4, P5 and P6 had the largest influence on predictions. The authors also report that the model predicts N501Y to broadly increase CD8+ T-cell immunogenicity, while the effects of D614G and E484K depend on position and HLA allele.
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 DeepImmuno on Tamarind Bio
Neoantigen prioritization: rank candidate tumor neoantigens by predicted immunogenicity.
Viral epitope screening: assess SARS-CoV-2 peptides and the effect of viral variants on T-cell immunogenicity.
Vaccine design: shortlist peptides likely to trigger CD8+ T-cell responses.
Peptide generation: explore new immunogenic-like sequences with the GAN component, if exposed.
How to Use DeepImmuno on Tamarind Bio
Log in and open the tool: sign in at tamarind.bio and open DeepImmuno.
Provide peptides: enter 9-mer or 10-mer peptide sequences, one per query or in a batch if supported.
Select the HLA allele: choose the MHC class I allele for each peptide.
Run the prediction: submit the job.
Download the outputs: retrieve the immunogenicity scores for each peptide-HLA pair.
Interpret the scores: the authors treated scores of 0.5 or higher as immunogenic; rank peptides rather than relying on a single cutoff.
Use downstream: combine with MHC binding prediction and experimental T-cell assays to confirm candidates.
Things to Keep in Mind
The authors state that specificity is low, with a 0.12 false-positive rate at the training-set level.
Sparse experimental evidence can inflate scores for some peptides.
TCR sequence information is not included because matched TCR data are scarce.
The model cannot distinguish HLA-bound from unbound peptides.
Neoantigen responses vary between patients, and generalization is uncertain given small validation sets such as 100 SARS-CoV-2 peptides.
Source: Li G, Iyer B, Prasath VBS, Ni Y, Salomonis N. DeepImmuno: deep learning-empowered prediction and generation of immunogenic peptides for T-cell immunity. Briefings in Bioinformatics 22(6), bbab160, 2021. DOI 10.1093/bib/bbab160. Code: https://github.com/frankligy/DeepImmuno. Web portal: https://deepimmuno.research.cchmc.org.