mber - SKILL.md Agent Skill

name: mber description: > VHH nanobody design using mBER (Manifold Binder Engineering and Refinement). Use this skill when: (1) Designing VHH nanobody CDRs against a target protein, (2) Have an existing VHH scaffold and want to redesign CDR1/CDR2/CDR3, (3) Optimizing a known VHH binder, (4) Targeting specific hotspot residues on the antigen, (5) Multi-chain antigen targets (with chain offsets).

For de novo antibody/nanobody CDR design, use iggm. For general protein binder design, use boltzgen or bindcraft. license: MIT category: design-tools tags: [antibody, nanobody, vhh, sequence-design, mask-based] source: https://github.com/hgbrian/biomodals

mBER VHH Nanobody Design

mBER uses structure templates and sequence conditioning (via AlphaFold-Multimer) to design VHH nanobody binders against a target protein.

Prerequisites

Requirement	Minimum	Recommended
GPU VRAM	40GB	80GB (A100)

How to run

Basic VHH design

# Design VHH against a PDB target (downloads PDB automatically)
modal run modal_mber.py \
  --target-pdb 7STF.pdb \
  --target-name PDL1

# Or by PDB ID
wget https://files.rcsb.org/download/7STF.pdb
modal run modal_mber.py --target-pdb 7STF.pdb --target-name PDL1

Custom masked VHH sequence (* = positions to design)

modal run modal_mber.py \
  --target-pdb target.pdb \
  --target-name MyTarget \
  --masked-binder-seq "EVQLVESGGGLVQPGGSLRLSCAASG*********WFRQAPGKEREF***********NADSVKGRFTISRDNAKNTLYLQMNSLRAEDTAVYYC************WGQGTLVTVSS"

The * characters specify which positions to design (typically CDR1, CDR2, CDR3). Framework regions remain fixed.

With specific chains and hotspots

modal run modal_mber.py \
  --target-pdb target.pdb \
  --target-name MyTarget \
  --chains A,B \
  --target-hotspot-residues A110,B120

Multi-chain with non-contiguous numbering

# Use chain-offsets to prevent folding chains as single chain
modal run modal_mber.py \
  --target-pdb target.pdb \
  --target-name MyTarget \
  --chains A,B \
  --chain-offsets B:200 \
  --target-hotspot-residues A6

Key parameters

Parameter	Default	Description
`--target-pdb`	required	Target PDB file
`--target-name`	required	Name for output files
`--masked-binder-seq`	default CDRs	VHH sequence with `*` for design positions
`--chains`	A	Target chains to include
`--target-hotspot-residues`	None	Residues to target (e.g., `A110,B120`)
`--chain-offsets`	None	Offset for multi-chain numbering (e.g., `B:200`)
`--output-dir`	`./out/mber`	Output directory

Default design regions

By default, mBER designs the three CDR loops of the VHH:

CDR1: positions ~26-35
CDR2: positions ~50-58
CDR3: positions ~97-110

The framework (FR1, FR2, FR3, FR4) is kept fixed.

Output format

out/mber/
├── design_0.pdb       # VHH-target complex
├── design_0.fasta     # Designed VHH sequence
└── scores.json        # AF-Multimer confidence scores

mBER vs IgGM

Aspect	mBER	IgGM
Input	Masked VHH sequence	X-masked FASTA
Method	AF-Multimer conditioning	Generative model
Antibody support	No (VHH only)	Yes (H+L)
Best for	VHH scaffold refinement	De novo CDR design

Troubleshooting

Error	Cause	Fix
`CUDA out of memory`	Large target	Use A100-80GB
`Chain not found`	Wrong chain ID	Check PDB chain IDs
`Invalid masked sequence`	Wrong length	Match VHH framework length

Next: Validate with chai or protenix → protein-qc for filtering.