Skip to content

README.md

Latest commit

 

History

History
87 lines (53 loc) · 4.39 KB

README.md

File metadata and controls

87 lines (53 loc) · 4.39 KB

Target Data

This directory contains PDB files used as design targets, organized by task type into subdirectories matching the source field in target/task config YAMLs.

The pipeline resolves target PDBs via $DATA_PATH/target_data/<source>/<target_filename>.pdb, so place your files accordingly or use an explicit target_path in the config.

Directory Structure

target_data/
├── bindcraft_targets/       # Protein binder targets (PD1, PDL1, CD45, etc.)
├── alpha_proteo_targets/    # AlphaProteo benchmark binder targets
├── ligand_targets/          # Ligand-only binder targets (small molecule + de novo protein)
└── ame_input_structures/    # AME (motif + ligand) scaffolding targets

Target Types

Protein Binder Targets (bindcraft_targets)

Standard protein-protein binder design. Each entry specifies a target chain, hotspot residues, and binder length range.

  • Config: configs/targets/targets_dict.yaml
  • Required fields: source, target_filename, target_input, hotspot_residues, binder_length

Ligand Binder Targets (ligand_targets)

Design a protein binder around a small-molecule ligand. The target PDB contains only the ligand; the protein is generated de novo.

  • Config: configs/targets/ligand_targets_dict.yaml
  • Additional fields: ligand (CCD code), SMILES, ligand_only: True, use_bonds_from_file: True

AME Targets (ame_targets)

Motif scaffolding with ligands (Atomistic Motif Extension). The target PDB contains a protein active site plus one or more ligands; the model scaffolds a new protein around the specified motif residues and ligand atoms.

  • Config: configs/design_tasks/ame_dict_v2.yaml
  • Additional fields: ligand, contig_atoms

Monomer Motif Targets

Motif scaffolding without ligands. Motif PDBs are referenced via $DATA_PATH/motif_data_aa/.

  • Config: configs/design_tasks/motif_dict.yaml

Preparing AME Target PDBs

Warning: AME input PDBs require manual cleaning before use. Most bundled structures in ame_input_structures/ still need preparation. M0024_1nzy_v3.pdb is provided as a ready-to-use example of a correctly cleaned structure. Malformed inputs will cause silent evaluation errors or incorrect RMSD values.

Required cleaning steps

  1. Chain assignment: Ligand(s) must be on chain A, motif protein residues on chain B. If your source PDB has different chain assignments, re-chain accordingly.

  2. Residue naming: Set the ligand residue name to L:0. This prevents RF3 from reconstructing missing atoms from the CCD dictionary, which causes shape mismatches in RMSD computation.

  3. Remove metal ions (Zn, Mg, Fe, etc.): Strip all metal/ion atoms from the PDB. These are sometimes included by accident from the source structure and will interfere with generation and evaluation. In the task config, also remove any metal CCD codes from the ligand list (e.g., remove "ZN", "MG", "FE", "FE2", "MN").

  4. Update contig_atoms: Make sure the chain letters in contig_atoms match the re-chained structure (should reference chain B for motif residues).

Example: cleaning with atomworks

from atomworks.io import load_any, to_pdb_file

atom_array = load_any("M0024_1nzy.pdb")[0]

# Remove metal ions
metal_names = {"ZN"} #, "MG", "FE", "MN", "CA", "NA", "CU", "CO", "NI"}
metal_mask = [rn.strip() in metal_names for rn in atom_array.res_name]
atom_array = atom_array[~numpy.array(metal_mask)]

# Re-chain: ligand -> A, protein -> B
atom_array.chain_id[atom_array.hetero] = "A"
atom_array.chain_id[~atom_array.hetero] = "B"

# Rename ligand residues to L:0
ligand_mask = atom_array.chain_id == "A"
atom_array.res_name[ligand_mask] = "L:0"

to_pdb_file(atom_array, "M0024_1nzy_v3.pdb")

Ligand naming for RF3 evaluation

RoseTTAFold3 (>=0.1.12, March 2026) reconstructs ligands from their CCD (Chemical Component Dictionary) code, adding back any atoms missing in the input PDB. If your ligand is incomplete (i.e., a substructure of the full CCD entry), RF3 will add the missing heavy atoms, causing a shape mismatch that breaks RMSD computation.

Setting ligand: "L:0" in the task config tells the pipeline to treat the ligand as a generic residue rather than resolving it from CCD, preserving the exact atom set from your input PDB. The M0024_1nzy_v3 entry in ame_dict_v2.yaml and its corresponding M0024_1nzy_v3.pdb are provided as a fully cleaned, ready-to-use reference.