Use Antibody Evolution (Affinity Maturation) Online
Commercially Available Antibody Evolution (Affinity Maturation) No-Code Web Server
Antibody Evolution: Affinity Maturation with General Protein Language Models
Language-model-guided antibody evolution is an approach that uses general protein language models to suggest affinity-improving mutations from a single antibody sequence. It was reported by Brian L. Hie, Varun R. Shanker, Duo Xu, Theodora U. J. Bruun, Payton A. Weidenbacher, Shaogeng Tang, Wesley Wu, John E. Pak and colleagues, with Peter S. Kim as corresponding author, in Nature Biotechnology (2023, open access).
Affinity maturation normally relies on large mutational libraries, screening and many experimental rounds. Computational shortcuts usually need antigen structures, task-specific training data or related sequences. This work asks whether models trained only on natural protein sequences already encode which substitutions are evolutionarily plausible, and whether that is enough to improve real antibodies.
The method needs only the wild-type VH or VL amino acid sequence. It uses no antigen information, no structure and no homologs, and it scores each single-residue substitution with a consensus of language models.
Property | Detail |
|---|---|
Input | Wild-type VH or VL amino acid sequence and a consensus cutoff |
Models | ESM-1b plus an ensemble of five ESM-1v models (six in total) |
Selection rule | Substitutions more likely than wild-type across the model consensus |
Output | About 10 recommended single substitutions, then combinations |
Speed | Under 1 s per antibody on GPU; 742 antibodies in about 3 minutes |
Antibodies tested | Seven, across influenza, Ebola and SARS-CoV-2 targets |
How Language-Model Affinity Maturation Works
Score substitutions: Each single-residue change is scored by the language models.
Take a consensus: Only substitutions favored over wild-type across the models are kept.
Round one: A small set of single substitutions is expressed and tested.
Round two: Beneficial substitutions are recombined and tested again.
Results: Affinity Gains Across Seven Antibodies
Measure | Result |
|---|---|
Variants expressed | All but one of 122 variants across seven trajectories |
Round-one Fabs keeping sub-micromolar binding | 71 to 100% |
Round-one variants improved at least 1.1-fold | 14 to 71% |
Matured antibodies | Up to about sevenfold improvement |
Unmatured antibodies | Up to 160-fold (mAb114 ancestor) |
Neutralization (IC50) | 32-fold for C143 vs Beta pseudovirus |
Changes in binding affinity tracked changes in neutralization (Spearman r = 0.82, n = 15). Of 31 affinity-enhancing variants, 21 had higher melting temperature than wild-type and all stayed above 70 C. No substantial change in polyspecificity was seen. General language models recommended more affinity-enhancing substitutions than the other methods compared, including antibody-specific models.
What is Tamarind Bio?
Tamarind Bio is a no-code bioinformatics platform built to give life scientists and researchers access to powerful computational tools. Many cutting-edge machine learning models are hard to deploy and use. Tamarind provides an intuitive, web-based environment that removes the complexity of high-performance computing, software dependencies and command-line interfaces.
The platform is designed for biologists, chemists and other researchers who may not have a background in programming or cloud infrastructure but want to run models on their own data. Key features include:
A user-friendly graphical interface for setting up and launching experiments
A robust API for integration into existing research pipelines
An automated system for managing and scaling computational resources
Tamarind treats information and data security as a top priority, as detailed in its Trust Center and Terms of Service.
Accelerating Discovery with Antibody Evolution on Tamarind Bio
Lead optimization: Improve the affinity of a therapeutic antibody candidate with a handful of tested variants.
Maturing early antibodies: Unmatured antibodies showed the largest gains, up to 160-fold in the paper.
Keeping up with variants: The authors improved neutralization of SARS-CoV-2 variant pseudoviruses.
Library design: Use the ranked substitutions to focus experimental libraries.
How to Use Antibody Evolution on Tamarind Bio
Open the tool: Log in at tamarind.bio and open Antibody Evolution.
Provide the sequence: Enter the wild-type VH or VL amino acid sequence.
Set the consensus cutoff: The paper uses a consensus across models with a cutoff parameter, if exposed.
Run the job: Launch the language model scoring.
Download suggestions: Retrieve the recommended single-residue substitutions.
Select and combine: Test a small set first, then combine the beneficial ones, as in the two-round design of the paper.
Validate: Express variants and measure binding, stability and neutralization.
Things to Keep in Mind
Only seven antibodies were tested, and gains on highly matured antibodies were modest (up to about sevenfold).
The method improves existing binders and does not confer new function.
Some variants, including certain MEDI8852 variants, had lower thermostability.
About half of the recommended substitutions were in framework regions, and their mechanisms are not fully explained.
Immunogenicity was only assessed with linear-peptide MHC binding predictions.
Source: Hie BL, Shanker VR, Xu D, Bruun TUJ, Weidenbacher PA, Tang S, Wu W, Pak JE, et al. Efficient evolution of human antibodies from general protein language models. Nature Biotechnology 2023 (vol. 42, pp. 275-283). DOI: 10.1038/s41587-023-01763-2. Article: nature.com/articles/s41587-023-01763-2.