Multi-resolution humanness scoring of antibodies via a region-isolated language model
- 1School of Pharmacy, Faculty of Medicine & Laboratory of Drug Discovery from Natural Resources and Industrialization, Macau University of Science and Technology,
Abstract
Antibodies recognized as foreign can elicit anti-drug antibodies that compromise efficacy and safety, making humanness scoring, estimating how human-like a sequence appears, an early, repeated step in antibody development. Widely used scores reduce a variable domain to a single framework-dominated number, leaving the complementarity-determining regions (CDRs) grafted during humanization, diversified by generative models weakly supervised and rarely audited. Here we introduce IgScope, a 23-million-parameter antibody language model trained from scratch that scores humanness at chain, IMGT-region, and residue resolution; a region-isolated attention module confines attention to within-region pairs, and chimeric-sequence training supplies region-level supervision. On 2.38 million held-out chains, masking all framework tokens leaves CDR accuracy unchanged, and back-mutated murine residues in human frameworks are detected at 93.4% accuracy. At chain level, IgScope matches the strongest published correlation with clinical anti-drug-antibody incidence (r = 0.567, n = 217) and outperforms a same-protocol recomputation. Region-level scoring exposes what chain-level scoring conceals: antibodies designed de novo by four generative models pass whole-chain humanness (0.966-0.988), while their CDRs score far below natural (P < 1e-30) in every chain class. Quality control turns the same resolution to the mirror-image question: in humanized-mouse platforms, whose domains are human by design, IgScope scores the domain as human and, in HUGO-Nano chimeric VHHs, isolates only the FR2 signal of four camelid hallmark residues deliberately retained to stabilize the VHH. IgScope gives antibody humanization a verified foundation and generative design an audit.
Related
- Moral echoes: do morality and humanness affect emotional and aesthetic responses to music?
- Pediatric Sleep Assessment through Spectral Sleep Scoring: A Comparison with Polysomnography
- Assessing accelerometry as an alternative to electromyography for mouse sleep scoring
- A foundation model learns the sequence and functional grammar of fully human heavy-chain-only antibodies
- HLA allele DRB1*01:03: the genetic link to neutralizing IL-10 antibodies in inflammatory bowel disease
BibTeX
@misc{liu2026multiresolution,
title = {{Multi-resolution humanness scoring of antibodies via a region-isolated language model}},
author = {H. Liu and P. Chen and S. Zhou and B. Wei and S. Deng and L. Zhang and X. Liu and S. Zhou and L. Han and L. Ye},
howpublished = {bioRxiv},
year = {2026},
month = oct,
doi = {10.64898/2026.10.23.761395},
url = {https://www.biorxiv.org/content/10.64898/2026.10.23.761395v1},
note = {CC BY-NC},
}
RIS
TY - UNPB
TI - Multi-resolution humanness scoring of antibodies via a region-isolated language model
AU - H. Liu
AU - P. Chen
AU - S. Zhou
AU - B. Wei
AU - S. Deng
AU - L. Zhang
AU - X. Liu
AU - S. Zhou
AU - L. Han
AU - L. Ye
PY - 2026
DA - 2026/10/23
PB - bioRxiv
DO - 10.64898/2026.10.23.761395
UR - https://www.biorxiv.org/content/10.64898/2026.10.23.761395v1
AB - Antibodies recognized as foreign can elicit anti-drug antibodies that compromise efficacy and safety, making humanness scoring, estimating how human-like a sequence appears, an early, repeated step in antibody development. Widely used scores reduce a variable domain to a single framework-dominated number, leaving the complementarity-determining regions (CDRs) grafted during humanization, diversified by generative models weakly supervised and rarely audited. Here we introduce IgScope, a 23-million-parameter antibody language model trained from scratch that scores humanness at chain, IMGT-region, and residue resolution; a region-isolated attention module confines attention to within-region pairs, and chimeric-sequence training supplies region-level supervision. On 2.38 million held-out chains, masking all framework tokens leaves CDR accuracy unchanged, and back-mutated murine residues in human frameworks are detected at 93.4% accuracy. At chain level, IgScope matches the strongest published correlation with clinical anti-drug-antibody incidence (r = 0.567, n = 217) and outperforms a same-protocol recomputation. Region-level scoring exposes what chain-level scoring conceals: antibodies designed de novo by four generative models pass whole-chain humanness (0.966-0.988), while their CDRs score far below natural (P < 1e-30) in every chain class. Quality control turns the same resolution to the mirror-image question: in humanized-mouse platforms, whose domains are human by design, IgScope scores the domain as human and, in HUGO-Nano chimeric VHHs, isolates only the FR2 signal of four camelid hallmark residues deliberately retained to stabilize the VHH. IgScope gives antibody humanization a verified foundation and generative design an audit.
ER -