Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2025 Dec;17(1):2534626.
doi: 10.1080/19420862.2025.2534626. Epub 2025 Jul 22.

AlphaBind, a domain-specific model to predict and optimize antibody-antigen binding affinity

Aditya A Agarwal  1 , James Harrang  1 , David Noble  1 , Kerry L McGowan  1 , Adrian W Lange  1 , Emily Engelhart  1 , Miranda C Lahman  1 , Jeffrey Adamo  1 , Xin Yu  2 , Oliver Serang  1 , Kyle J Minch  1 , Kimberly Y Wellman  1 , David A Younger  1 , Randolph M Lopez  1 , Ryan O Emerson  1
Affiliations

AlphaBind, a domain-specific model to predict and optimize antibody-antigen binding affinity

Aditya A Agarwal et al. MAbs. 2025 Dec.

Abstract

Antibodies are versatile therapeutic molecules that use combinatorial sequence diversity to cover a vast fitness landscape. Designing optimal antibody sequences, however, remains a major challenge. Recent advances in deep learning provide opportunities to address this challenge by learning sequence-function relationships to accurately predict fitness landscapes. These models enable efficient in silico prescreening and optimization of antibody candidates. By focusing experimental efforts on the most promising candidates guided by deep learning predictions, antibodies with optimal properties can be designed more quickly and effectively. Here we present AlphaBind, a domain-specific model that uses protein language model embeddings and pre-training on millions of quantitative laboratory measurements of antibody-antigen binding strength to achieve state-of-the-art performance for guided affinity optimization of parental antibodies. We demonstrate that an AlphaBind-powered antibody optimization pipeline can deliver candidates with substantially improved binding affinity across four parental antibodies (some of which were already affinity-matured) and using two different types of training data. The resulting candidates, which include up to 11 mutations from parental sequence, yield a sequence diversity that allows optimization of other biophysical characteristics, all while using only a single round of data generation for each parental antibody. AlphaBind weights and code are publicly available at: https://github.com/A-Alpha-Bio/alphabind.

Keywords: Antibody engineering; computational protein design; machine learning; yeast display.

PubMed Disclaimer

Conflict of interest statement

All A-Alpha Bio-affiliated authors were employees of A-Alpha Bio, Inc. (A-Alpha Bio) at the time the research was performed, and own stock/stock options of A-Alpha Bio. A-Alpha Bio has a patent application relating to certain research described in this article.

Figures

Diagram showing AlphaBind’s end-to-end process. Panel A presents three parental antibodies used in this study. Panel B outlines pre-training and fine-tuning steps, sequence optimization, and experimental validation using high-throughput yeast display and BLI to assess binding affinity improvements.
Figure 1.
Experimental overview.
Schematic of AlphaBind’s architecture. Panel A shows the ESM2-nv embedding step for antibody and target, and the transformer model used for affinity prediction. Panel B explains the iterative optimization method that selects improved variants using masked sampling and affinity prediction. Panel C outlines the full process—training, optimization, and selection for predicted affinity and predicted developability.
Figure 2.
AlphaBind Architecture.
Bar charts illustrate per-residue sequence diversity (as per-residue Shannon entropy) across all optimized variants of three parental antibodies. AlphaBind proposes mutations across a wide range of positions, including CDRs and framework regions, showing diverse strategies for affinity optimization beyond just CDRs.
Figure 3.
Sequence diversity of optimized candidates.
Violin plots of AlphaSeq affinities at each edit distance from two through eleven (Panel A) show improved affinity for AlphaBind-generated variants across all tested edit distances. Panel B presents BLI results, confirming that all predicted top candidates from the three systems demonstrated superior or comparable binding compared to their parental antibodies.
Figure 4.
AlphaBind validation results.
Optimization of anti-TIGIT antibody AAB-PP3115 for developability. Panel A highlights non-germline residues and sequence liabilities. Panel B shows edited residues, predicted affinity, and BLI affinity for 11 top AlphaBind-generated variants with ablated liabilities and germline reversions, which additionally retain or improve binding affinity using BLI despite up to nine additional mutations.
Figure 5.
Affinity-guided developability engineering.
Line plots comparing median candidate affinity across model variants. Left: among all candidates, the full model significantly outperforms versions lacking pre-training or ESM-2nv embeddings. Right: analysis of top-performing 10% of candidates confirms similar trends, showing both architectural features are essential for optimal candidate design.
Figure 6.
AlphaBind ablation analysis.
AlphaBind optimization of Trastuzumab-scFv using external DMS data, and optimized in two ways: CDRH3-only and full CDRH2–FR3–CDRH3. Panel A shows AlphaSeq affinity results, Panel B shows sequence diversity, and Panel C confirms expression and binding of top candidates via BLI.
Figure 7.
Optimization of Trastuzumab-scFv.
See this image and copyright information in PMC

Similar articles

See all similar articles

References

    1. Lu RM, Hwang Y-C, Liu I-J, Lee C-C, Tsai H-Z, Li H-J, Wu H-C.. Development of therapeutic antibodies for the treatment of diseases. J Biomed Sci. 2020;27(1). doi: 10.1186/s12929-019-0592-z. - DOI - PMC - PubMed
    1. Wang Z, Wang G, Lu H, Li H, Tang M, Tong A. Development of therapeutic antibodies for the treatment of diseases. Mol Biomed. 2022;3(35). doi: 10.1186/s43556-022-00100-4. - DOI - PMC - PubMed
    1. Bashour H, Smorodina E, Pariset M, Zhong J, Akbar R, Chernigovskaya M, Lê Quý K, Snapkow I, Rawat P, Krawczyk K, et al. Biophysical cartography of the native and human-engineered antibody landscapes quantifies the plasticity of antibody developability. Commun Biol. 2024;7(1):922. doi: 10.1038/s42003-024-06561-3. - DOI - PMC - PubMed
    1. Jain T, Sun T, Durand S, Hall A, Houston NR, Nett JH, Sharkey B, Bobrowicz B, Caffry I, Yu Y. et al. Biophysical properties of the clinical-stage antibody landscape. Proc Natl Acad Sci USA. 2017;114(5):944–17. doi: 10.1073/pnas.1616408114. - DOI - PMC - PubMed
    1. Raybould MIJ, Deane CM. The therapeutic antibody profiler for computational developability assessment. Methods Mol Biol. 2022;2313:115–125. doi: 10.1007/978-1-0716-1450-1_5. - DOI - PubMed

MeSH terms

LinkOut - more resources