MolecularDiffusion.data.component.diffpharma_prep

Build a DiffPharma complex from a raw protein PDB + reference ligand SDF.

Port of the only working novel-pocket path in the upstream repo: test_single.py’s process_data + process_data_h, i.e. process_crossdock.process_ligand_and_pocket (full-atom branch) plus interaction_construct.{hbond_create,hydrophobic_data}.

Not ported: the ca_only=True pocket branch (needs Bio.PDB.Polypeptide.three_to_one, removed in Biopython >= 1.80, and the released weights are full-atom anyway), and lightning_modules.py:921 generate_ligands (stale DiffSBDD leftover that calls sample_given_pocket with a pre-DiffPharma signature).

“Novel pocket” here means “novel protein + a reference ligand pose”: the SDF is both the 8 A pocket-selection reference and the ligand ODDT detects the protein-ligand interactions against. That is upstream’s own framing.

Heavy dependencies (Biopython, ODDT/OpenBabel, RDKit) are imported inside the functions and gated by optional.require_modules("bio", ...), so importing this module never fails on a bare install.

Attributes

ATOM_DICT
HYDROPHOBIC_TYPES

Functions

complex_from_files(→ Dict[str, Any])	PDB + reference SDF -> the same per-complex dict a converted db row yields.
hbond_particles(→ Tuple[numpy.ndarray, numpy.ndarray, ...)	3 pseudo-nodes per H-bond, at 1/4, 1/2 and 3/4 along it.
hydrophobic_particles(→ Tuple[numpy.ndarray, ...)	2 pseudo-nodes per hydrophobic contact, at 1/3 and 2/3 along it.
process_ligand_and_pocket(→ Tuple[Dict[str, ...)	Ligand atoms + every atom of each residue within dist_cutoff of it.

Module Contents

MolecularDiffusion.data.component.diffpharma_prep.complex_from_files(pdb_file: str, sdf_file: str, center: bool = False, dist_cutoff: float = 8.0) → Dict[str, Any]

PDB + reference SDF -> the same per-complex dict a converted db row yields.

MolecularDiffusion.data.component.diffpharma_prep.hbond_particles(protein, ligand) → Tuple[numpy.ndarray, numpy.ndarray, numpy.ndarray]

3 pseudo-nodes per H-bond, at 1/4, 1/2 and 3/4 along it.

Classes are [SP, DD, AC] tiled [0, 1, 1] for protein-donor bonds and [0, 2, 2] for protein-acceptor bonds.

MolecularDiffusion.data.component.diffpharma_prep.hydrophobic_particles(protein, ligand) → Tuple[numpy.ndarray, numpy.ndarray, numpy.ndarray]

2 pseudo-nodes per hydrophobic contact, at 1/3 and 2/3 along it.

MolecularDiffusion.data.component.diffpharma_prep.process_ligand_and_pocket(pdbfile: str, sdffile: str, dist_cutoff: float = 8.0) → Tuple[Dict[str, numpy.ndarray], Dict[str, numpy.ndarray]]

Ligand atoms + every atom of each residue within dist_cutoff of it.

MolecularDiffusion.data.component.diffpharma_prep.ATOM_DICT

MolecularDiffusion.data.component.diffpharma_prep.HYDROPHOBIC_TYPES