The Quagmire of Immunogenicity-Guided Antibody Design: Labels Limit Progress
- 1Leiden University Medical Center
Abstract
Generative sequence-based antibody design is increasingly steered by learned property predictors, so a design objective is only as well specified as the labels behind its oracle. Immunogenicity is among the most clinically consequential properties to condition on, and among the least well specified: the supervisory signal is a single anti-drug-antibody percentage (ADA%) per molecule, aggregated from published trials. To analyze that signal, we first extended and corrected the widely used 217-entry dataset to 231 therapeutics with paired sequences, spanning formats from mouse to fully human. Then, we manually reconstructed the clinical evidence for 110 of them from primary sources. We find that an antibody format lookup outperforms every sequence-based learned predictor on the 231-antibody dataset. Apparent accuracy tracks the format composition of the evaluation split rather than sequence signal. Furthermore, the labels understate their own evidence: molecules labeled at [≤]6 ADA% span up to 86% across trials. We recomputed labels from the trial-level measurements and found that the advantage of the format-based approach disappears, while the sequence-based predictors degrade less under the cohort-weighted mean. We conclude that progress is limited by the label itself, and recommend reporting ADA% at the trial level, treating the target as a distribution rather than a point value, and evaluating against a format-based baseline.
Related
- AbRefine: Framework-Conditioned Refinement for Structure-Guided Antibody CDR Design
- Clinical immunogenicity assessment of HIV/AIDS vaccines: current progress and future perspectives.
- Mechanisms of antibody-dependent enhancement and its research progress in animal coronaviruses
- Function-preserving CD8⁺ T Cell Immunogenicity minimization through protein design
- Comprehensive evaluation of antibody language models across multiple antibody design tasks
BibTeX
@misc{villegasmorcillo2026quagmire,
title = {{The Quagmire of Immunogenicity-Guided Antibody Design: Labels Limit Progress}},
author = {A. Villegas-Morcillo and C. Raman and J. M. Weber and M. J. T. Reinders},
howpublished = {bioRxiv},
year = {2026},
month = oct,
doi = {10.64898/2026.10.27.762175},
url = {https://www.biorxiv.org/content/10.64898/2026.10.27.762175v1},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - The Quagmire of Immunogenicity-Guided Antibody Design: Labels Limit Progress
AU - A. Villegas-Morcillo
AU - C. Raman
AU - J. M. Weber
AU - M. J. T. Reinders
PY - 2026
DA - 2026/10/27
PB - bioRxiv
DO - 10.64898/2026.10.27.762175
UR - https://www.biorxiv.org/content/10.64898/2026.10.27.762175v1
AB - Generative sequence-based antibody design is increasingly steered by learned property predictors, so a design objective is only as well specified as the labels behind its oracle. Immunogenicity is among the most clinically consequential properties to condition on, and among the least well specified: the supervisory signal is a single anti-drug-antibody percentage (ADA%) per molecule, aggregated from published trials. To analyze that signal, we first extended and corrected the widely used 217-entry dataset to 231 therapeutics with paired sequences, spanning formats from mouse to fully human. Then, we manually reconstructed the clinical evidence for 110 of them from primary sources. We find that an antibody format lookup outperforms every sequence-based learned predictor on the 231-antibody dataset. Apparent accuracy tracks the format composition of the evaluation split rather than sequence signal. Furthermore, the labels understate their own evidence: molecules labeled at [≤]6 ADA% span up to 86% across trials. We recomputed labels from the trial-level measurements and found that the advantage of the format-based approach disappears, while the sequence-based predictors degrade less under the cohort-weighted mean. We conclude that progress is limited by the label itself, and recommend reporting ADA% at the trial level, treating the target as a distribution rather than a point value, and evaluating against a format-based baseline.
ER -