{"id":"arxiv:2512.17815","kind":"preprint","title":"Structure-Aware Antibody Design with Affinity-Optimized Inverse Folding","authors":["Xinyan Zhao","Yi-Ching Tang","Rivaaj Monsia","Victor J. Cantu","Ashwin Kumar Ramesh","Xiaozhong Liu","Zhiqiang An","Xiaoqian Jiang","Yejin Kim"],"author_count":9,"year":2025,"published":"2025-12-19","venue":null,"cited_by":0,"doi":null,"arxiv_id":"2512.17815","nct_id":null,"source_url":"https://arxiv.org/abs/2512.17815","archive_url":"https://space-kitty.com/papers/arxiv:2512.17815","updated":"2025-12-19","pmid":null,"pmcid":null,"openalex_id":"W7116925460","pdf_url":"https://arxiv.org/pdf/2512.17815","abstract":"Motivation: The clinical efficacy of antibody therapeutics critically depends on high-affinity target engagement, yet laboratory affinity-maturation campaigns are slow and costly. In computational settings, most protein language models (PLMs) are not trained to favor high-affinity antibodies, and existing preference optimization approaches introduce substantial computational overhead without clear affinity gains. Therefore, this work proposes SimBinder-IF, which converts the inverse folding model ESM-IF into an antibody sequence generator by freezing its structure encoder and training only its decoder to prefer experimentally stronger binders through preference optimization. Results: On the 11-assay AbBiBench benchmark, SimBinder-IF achieves a 55 percent relative improvement in mean Spearman correlation between log-likelihood scores and experimentally measured binding affinity compared to vanilla ESM-IF (from 0.264 to 0.410). In zero-shot generalization across four unseen antigen-antibody complexes, the correlation improves by 156 percent (from 0.115 to 0.294). SimBinder-IF also outperforms baselines in top-10 precision for ten-fold or greater affinity improvements. A case study redesigning antibody F045-092 for A/California/04/2009 (pdmH1N1) shows that SimBinder-IF proposes variants with substantially lower predicted binding free energy changes than ESM-IF (mean Delta Delta G -75.16 vs -46.57). Notably, SimBinder-IF trains only about 18 percent of the parameters of the full ESM-IF model, highlighting its parameter efficiency for high-affinity antibody generation.","abstract_license":"CC0-1.0 (arXiv descriptive metadata)","license":null,"is_open_access":false,"retracted":false,"peer_reviewed":false,"categories":["cs.CE","q-bio.BM"],"details":null,"provenance":"arxiv","attribution":"Abstract: arXiv descriptive metadata (CC0). Thank you to arXiv for use of its open access interoperability. This service was not reviewed or approved by, nor does it necessarily express or reflect the policies or opinions of, arXiv.","topics":[{"slug":"protein-design","name":"AI Protein Design & Structure Prediction","domain_slug":"biotech"}],"cited_in_dossiers":[]}