docs - ligandmpnn | subseq.bio

Inputs

Place structure inputs (PDB plus ligand files) under /inputs.
Batch mode: -I=<dir> (e.g. -I=/inputs) processes all input files in a folder.
Results and designed sequences are written to /outputs.

Arguments

--model_type: protein_mpnn, ligand_mpnn, per_residue_label_membrane_mpnn, global_label_membrane_mpnn, soluble_mpnn.
--checkpoint_protein_mpnn: path to ProteinMPNN weights.
--checkpoint_ligand_mpnn: path to LigandMPNN weights.
--checkpoint_per_residue_label_membrane_mpnn: path to per-residue membrane model weights.
--checkpoint_global_label_membrane_mpnn: path to global-label membrane model weights.
--checkpoint_soluble_mpnn: path to soluble model weights.
--fasta_seq_separation: separator between chains in FASTA output.
--verbose: print verbose logs.
--pdb_path: input PDB path.
--pdb_path_multi: JSON file listing PDB paths (keys used).
--fixed_residues: residues to keep fixed (e.g. A12 A13 B2).
--fixed_residues_multi: JSON mapping of fixed residues per PDB.
--redesigned_residues: residues to redesign (everything else fixed).
--redesigned_residues_multi: JSON mapping of redesigned residues per PDB.
--bias_AA: global amino-acid bias string (e.g. A:-1.0,P:2.0).
--bias_AA_per_residue: JSON mapping per-residue amino-acid biases.
--bias_AA_per_residue_multi: JSON mapping per-PDB per-residue biases.
--omit_AA: amino acids to omit globally (e.g. ACG).
--omit_AA_per_residue: JSON mapping per-residue omissions.
--omit_AA_per_residue_multi: JSON mapping per-PDB per-residue omissions.
--symmetry_residues: symmetry groups (e.g. A12,A13|C2,C3).
--symmetry_weights: weights corresponding to --symmetry_residues.
--homo_oligomer: set to 1 to auto-configure homo-oligomer symmetry.
--out_folder: output folder (use a path under /outputs).
--file_ending: suffix appended to output filenames.
--zero_indexed: set to 1 to start output numbering at 0.
--seed: RNG seed.
--batch_size: sequences per pass.
--number_of_batches: number of batches to sample.
--temperature: sampling temperature.
--save_stats: set to 1 to save output stats.
--ligand_mpnn_use_atom_context: set to 1 to use atom context.
--ligand_mpnn_cutoff_for_score: protein-context cutoff in angstroms.
--ligand_mpnn_use_side_chain_context: set to 1 to use side-chain context.
--chains_to_design: chains to redesign (comma-separated).
--parse_these_chains_only: chains to parse (comma-separated).
--transmembrane_buried: buried residues for per-residue membrane model.
--transmembrane_interface: interface residues for per-residue membrane model.
--global_transmembrane_label: global label for global membrane model.
--parse_atoms_with_zero_occupancy: set to 1 to parse zero-occupancy atoms.
--pack_side_chains: set to 1 to run side-chain packing.
--checkpoint_path_sc: side-chain packer weights.
--number_of_packs_per_design: packs per design.
--sc_num_denoising_steps: denoising steps for packing.
--sc_num_samples: samples per packing run.
--repack_everything: set to 1 to repack fixed residues too.
--force_hetatm: set to 1 to force ligand atoms as HETATM.
--packed_suffix: suffix for packed PDBs.
--pack_with_ligand_context: set to 1 to include ligand context during packing.

Example arguments

Defaults that mirror the New Job prefill:

--model_type=protein_mpnn
-I=/inputs
--out_folder=/outputs
--temperature=0.05
--checkpoint_protein_mpnn=/ref/model_params/proteinmpnn_v_48_020.pt

Tweak --temperature or swap checkpoints if you have custom weights.

Submit

Launch via New Job -> LigandMPNN.

Output notes

LigandMPNN produces a multi-FASTA output. The first record is the input sequence when available.
For backbone-only inputs, the “input sequence” is often all glycine (not useful for many downstream tools).
The designed (generated) sequence is often the second record and is typically what you want to extract/use downstream.