Use ImmunoSeq Online
Commercially Available ImmunoSeq No-Code Web Server
ImmunoSeq: Interpretable Antibody Immunogenicity and Humanness Scoring from Sequence
ImmunoSeq is an interpretable, sequence-based method that estimates the anti-drug antibody (ADA) risk of therapeutic antibodies by checking how many short peptides in a sequence match a library of human self-peptides. It also scores every residue, so you can see which positions look immunogenic and which framework mutations might lower risk. It was described in a 2025 bioRxiv preprint; the pages we could read did not list authors, affiliations, a code repository or a license.
Anti-drug antibodies can blunt the efficacy of biologics and cause adverse reactions, so predicting immunogenicity early in development matters. Existing approaches each leave a gap: MHC-binding predictors ignore downstream steps, T-cell epitope models depend on limited annotated data, homology-based methods lack residue-level resolution, and deep learning models are black boxes that give little guidance on which edits to make.
Attribute | Detail |
|---|---|
Core idea | Immune tolerance: sequences resembling human self-proteins should carry lower ADA risk |
Input | Antibody heavy and light chain sequences |
Fragmentation | Overlapping k-mers, k = 8 to 12 |
Score range | -0.2 (high immunogenicity) to 1 (high tolerance) |
Clinical benchmark | 217 antibodies with ADA data: Pearson R = -0.53, Spearman rho = -0.41 |
Humanness benchmark | 553 antibodies: ROC-AUC 0.982, MCC 0.910 |
How ImmunoSeq Scores Antibody Immunogenicity
Rather than training a black-box classifier, ImmunoSeq counts peptide matches against two libraries and combines them into a transparent score.
Fragment the sequence: each chain is split into overlapping k-mers of length 8 to 12.
Match against a self library: positive hits come from human proteins and healthy-donor human antibodies and add +1.0 each.
Penalize non-self matches: hits in a mouse antibody library subtract 0.2 each.
Normalize and combine: totals are normalized by peptide count; heavy and light chains are scored separately and averaged.
Profile residues: a residue's score is computed from all k-mers that contain it, flagging low-scoring residues as potential immunogenic hotspots.
The libraries are large: the authors report over 700 million positive and about 40 million negative peptides, built from 21,931 unique human proteins, 1,177,822 paired human antibodies and 84,449 paired mouse antibodies. Bloom filters keep lookups memory-efficient.
Validation: ADA Correlation, Humanness Classification and Humanization Design
Clinical ADA and humanness
Scores correlated negatively with clinical ADA incidence across 217 antibodies. Median hit rates were 0.83 for human and 0.34 for mouse antibodies, and antibodies scoring above 0.8 had median ADA below 5%, while those below 0.4 had median ADA above 40%.
Test | Result |
|---|---|
Humanness classification (198 human vs 355 non-human) | ROC-AUC 0.982, MCC 0.910 at threshold 0.75 |
PR-AUC | 0.959 (AbNatiV 0.971, OASis 0.953) |
Humanization pairs (25) | Humanized sequences scored higher in every case, averaging +0.3 |
Iterative greedy design | Median hit rate 0.84 vs 0.75 for experimental humanized sequences |
Nanobodies (8 SARS-CoV-2) | Hit rate gains of about 0.1, confined to framework regions |
Guiding mutations
Two design strategies were tested: a one-shot ranking of single-point framework mutations and an iterative greedy approach that adds the best mutation each round. One-shot precision at K=1 was 0.60 and greedy precision at K=5 reached 0.65, measured against experimental humanization mutations.
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 ImmunoSeq on Tamarind Bio
Early developability triage: rank candidate antibodies by predicted ADA risk before committing to expensive development.
Humanization guidance: identify low-scoring framework residues and test candidate substitutions.
Humanness assessment: classify sequences as human-like or non-human-like using the 0.75 threshold the authors report.
Nanobody engineering: explore framework changes that raise predicted tolerance while leaving CDRs unchanged.
How to Use ImmunoSeq on Tamarind Bio
Open the tool: log in to tamarind.bio and open ImmunoSeq.
Provide sequences: enter the antibody heavy and light chain sequences you want to score.
Set parameters: if exposed, review options such as k-mer length (8 to 12) and the design mode (one-shot or iterative).
Run the job: submit and wait for the scoring to finish.
Download outputs: save the overall score and the residue-level profile.
Interpret results: scores near 1 indicate high tolerance; low-scoring residues are candidate immunogenic hotspots.
Iterate: apply suggested framework mutations, rescore, and combine with your structure and developability checks.
Things to Keep in Mind
The method is sequence-only and ignores structural epitopes and surface features.
It uses exact-match peptide hits and does not weight protease cleavage.
The -0.2 penalty and 0.75 threshold were chosen empirically and may depend on library composition.
Greedy design can converge to local optima.
Human versus rhesus discrimination performed poorly for most models, including ImmunoSeq.
Experimental humanization mutations are not the only valid reference, and the nanobody test set was small (8).
Source: ImmunoSeq preprint, bioRxiv, 2025, DOI 10.1101/2025.08.14.670305, https://www.biorxiv.org/content/10.1101/2025.08.14.670305v1. Authors and a code repository were not available in the pages we read.