Use mBER Online
Commercially Available mBER No-Code Web Server
mBER: Open-Source De Novo VHH Nanobody Design with AlphaFold-Multimer Backpropagation
mBER (Manifold Binder Engineering and Refinement) is an open-source system for format-specific binder design, focused on antibody-format VHH binders, using backpropagation through AlphaFold-Multimer (AF2-M). The preprint describes it as built on ColabDesign and BindCraft.
General-purpose binder design methods are not specialized for antibody formats, whose frameworks are conserved and whose binding loops are constrained. mBER addresses this by giving AF2-M structural templates and sequence priors for VHHs, so that designs are partially constrained in both sequence and structure.
Notably, the authors report that this guidance is sufficient for backpropagation-based design of high-confidence multimeric structures, without additional model training or fine-tuning.
Property | Detail |
|---|---|
Task | De novo design of VHH binders against a target structure |
Folding model | AlphaFold-Multimer, optimized by backpropagation |
Sequence prior | ESM2 logits over masked framework positions |
Structural template | NanoBodyBuilder2 |
Inputs | Target structure, optional hotspot, partially masked VHH framework |
How mBER Designs VHH Binders with AlphaFold-Multimer
Prepare inputs: a target structure with an optional hotspot, plus a partially masked VHH framework sequence.
Truncate the target: the target is automatically cut to a smaller, foldable epitope region.
Sample a sequence prior: ESM2 produces logits over masked positions and one sequence is sampled.
Build a template: the sampled sequence is folded with NanoBodyBuilder2.
Optimize: template features and a continuous sequence representation go into AF2-M, and losses from its confidence module and structure drive iterative sequence optimization.
Score: a final forward pass returns the structure and confidence metrics, including ipTM and pLDDT.
mBER Experimental Results: Phage Display Validation Across Human Cell-Surface Targets
The authors designed 1,153,241 unique VHH binders against 436 human cell-surface proteins and screened designs for 145 targets by multiplexed phage display with NGS readout.
Metric | Reported value |
|---|---|
Targets with significant on-design binding | 65 of 145 (45%) |
Targets with any on-design binding | 78 of 145 |
Library 1 | 565,982 designs, 370 targets; 22 significant targets (36%) |
Library 2 (ipTM above 0.66) | 587,259 designs, 389 targets; 49 significant targets (42%) |
Median per-target hit rate | About 0.4% (range about 0.02% to 8%) |
Hit rate at ipTM 0.85-0.9 | 3.9% |
ipTM was a better predictor of on-design binding than pLDDT, and off-design hit rates stayed relatively flat across ipTM bins. Under strict filtering on favorable epitopes, hit rates reached up to 38% (8 of 21).
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 mBER on Tamarind Bio
Nanobody discovery: generate VHH binder candidates against a cell-surface target before wet-lab screening.
Epitope-focused design: direct designs toward a chosen hotspot on the target.
Library construction: produce large design sets for multiplexed phage display, as the authors did.
Candidate ranking: use ipTM filtering to enrich for designs more likely to bind.
How to Use mBER on Tamarind Bio
Log in and open the tool: sign in at tamarind.bio and select mBER.
Provide the target: upload or specify the target structure.
Choose a hotspot: if exposed, set the epitope residues, since hit rates depend strongly on the epitope.
Set the VHH framework: use the default partially masked framework unless the form exposes options.
Set the number of designs: the paper generated large numbers per target, so more samples help.
Run the job: submit and wait for AF2-M optimization to finish.
Filter by ipTM: download designs and prioritize high ipTM, for example above 0.7 or 0.8 as discussed in the paper.
Things to Keep in Mind
Successful designs cover only a subset of epitopes on a subset of targets.
Predicted binding modes were not structurally validated, and phage display does not precisely quantify affinity.
Deeper sampling improves ipTM and pLDDT but becomes expensive at scale.
Quality depends on AF2-M accuracy, and hyperparameters were largely inherited from ColabDesign and BindCraft without exhaustive tuning.
Source: mBER (Manifold Binder Engineering and Refinement) bioRxiv preprint (2025), doi.org/10.1101/2025.09.26.678877. The fetched page lists no code repository URL.