Skip to main content
mBER is an open-source protein design framework specifically engineered for antibody-format binder design. By leveraging structural templates and sequence conditioning within the ColabDesign framework, mBER enables backpropagation-based design through AlphaFold-Multimer to produce high-affinity VHH (nanobody) binders.

​
Design Philosophy

Unlike “minibinder” tools that generate entirely novel structures, mBER focuses on format-specific design. It maintains the highly conserved framework regions of antibodies while optimizing the variable regions (CDRs) for target binding.

​
Experimental Validation

mBER is one of the first open-source methods to demonstrate double-digit percentage experimental success rates for antibody design.
  • Scale: Validated through the screening of over 1 million VHH designs against 145 targets.
  • Success Rate: Achieved specific binding success against 45% of tested targets.
  • Precision: After filtering for high-confidence designs, hit rates for specific epitopes reached as high as 38%.

​
Methods

The mBER pipeline is built around three modular steps:
  1. Template Generation: Prepares the target structure (with optional truncation) and uses NanoBodyBuilder2 to create a structural VHH template.
  2. Trajectory Optimization: Designs binder sequences by iteratively optimizing a continuous sequence representation against AlphaFold-Multimer loss functions, guided by sequence priors from ESM2 or AbLang2.
  3. Evaluation: Final designs are re-folded and scored using AlphaFold-Multimer confidence metrics to identify the most promising candidates.

​
Filtering & Best Practices

The authors highlight that AlphaFold-Multimer’s ipTM score is the most reliable predictor of experimental success:
  • Recommended Filter: ipTM > 0.7 for a significant increase in hit rates.
  • Strict Filter: ipTM > 0.8 for the highest probability of finding specific binders.
  • Note: While pLDDT reflects the quality of the monomeric fold, it is a weaker predictor of actual binding success than ipTM.

​
Configuration (mber-vhh)

When running mBER on Tamarind, you can configure the following settings:
SettingTypeDescription
Target ProteinPDBThe structure of the protein you wish to bind to.
Target ChainsStringChain IDs of the target (e.g., A, B).
Hotspot ResiduesListSpecific epitope residues to target. If empty, mBER auto-detects sites.
VHH TemplateSeqFramework sequence with * marking CDR positions for design.
PLM ModelEnumThe language model used for guidance (ESM or Ablang2).
TemperatureFloatControls sequence sampling diversity (0.1 to 2.0).

​
Target Selection Tips

mBER performance is highly sensitive to the chosen epitope.
  • Success rates vary significantly between different hotspots on the same target.
  • Because mBER provides prospective epitope information, it is highly useful for designing binders that block specific protein-protein interactions.
Try mber