ligandmpnn-reference

LigandMPNN Reference Skill

Full documentation: docs/plans/2026-01-15-ligandmpnn-comprehensive-skill.md

Quick Reference

Model Types

Key Parameters

--model_type ligand_mpnn          # Model selection
--temperature 0.1                  # Sampling diversity (0.1 = default)
--fixed_residues "A10 A15 B20"    # Keep these positions unchanged
--bias_AA "H:5.0,D:3.0,E:3.0"     # Favor specific amino acids
--omit_AA "C,M"                    # Never use these amino acids
--ligand_mpnn_use_atom_context 1   # Enable ligand context
--pack_side_chains 1               # Enable side chain packing

Residue Format

A12       → Chain A, residue 12
B82A      → Chain B, residue 82, insertion code A
"A10 A15" → Space-separated list

Common Bias Presets

# Metal binding (Zn, Fe, Cu)
--bias_AA "H:5.0,C:4.0,D:3.0,E:3.0" --omit_AA "P,G"

# Lanthanide binding (Tb, Gd, Eu)
--bias_AA "D:6.0,E:4.0,N:1.0,Q:1.0" --omit_AA "C,H"

# Small molecule (aromatic)
--bias_AA "W:3.0,Y:2.0,F:2.0,H:1.5"

# Nucleotide binding
--bias_AA "R:4.0,K:3.0,N:2.0,Q:2.0,S:1.5,T:1.5"

Temperature Guide

Output Metrics

• overall_confidence: Average confidence (0-1)
• ligand_confidence: Near-ligand confidence
• seq_rec: Sequence recovery vs original

Example: Metal Site Design

python run.py \
    --model_type ligand_mpnn \
    --pdb_path ./protein_with_metal.pdb \
    --fixed_residues "A10 A15 A20 A25" \
    --bias_AA "D:5.0,E:4.0,H:3.0" \
    --ligand_mpnn_use_atom_context 1 \
    --temperature 0.1 \
    --out_folder ./output/

Best Practices

• Always fix catalytic/coordinating residues
• Use --ligand_mpnn_use_side_chain_context 1 for fixed residue context
• Lower temperature = higher confidence, less diversity
• Check ligand_confidence for binding site quality