ImmunoSeq

Methods to predict antibody immunogenicity

ImmunoSeq -- an interpretable and applicable method for immunogenicity prediction rooted in the biological principle of immune tolerance

Installation Guide

1. First check that you have installed python packages listed in requirements.txt.
2. Then download paired human and mouse antibody sequence files from Observed Antibody Sequences (https://opig.stats.ox.ac.uk/webapps/oas/oas_paired/) and move them into data/ folder
3. Run python prepare.py to generate k-mer (k=8-12) peptide library for human proteins, oas paired human antibodies, as well as oas paired mouse antibodies

Benchmark

1. To run ADA correlation benchmark, use python eval_ada_correlation.py
2. To run humanness classification benchmark, use python eval_humanness_classification.py
3. To benchmark humanness classification on anbativ dataset, run eval_abnativ.ipynb
4. To analyze Hu-mAb 25 antibody pairs, use python eval_humab25.py
5. To perform sequence immunogenicity optimization, use python infer.py

Citation

@article{bytedance2025ImmunoSeq,
  title={Antibody immunogenicity prediction and optimization with ImmunoSeq},
  author={Huang, Qiaojing and He, Yi and Liu, Kai},
  year={2025},
  journal={bioRxiv},
  publisher={Cold Spring Harbor Laboratory},
  doi={10.1101/2025.08.14.670305},
  URL={https://www.biorxiv.org/content/10.1101/2025.08.14.670305v1},
  elocation-id={2025.08.14.670305},
  eprint={https://www.biorxiv.org/content/10.1101/2025.08.14.670305v1.full.pdf},
}

Please address all questions to huangqiaojing@bytedance.com