Preprocessing

Stage 1 standardizes input molecules and rejects entries that are unsuitable for later stages. It defines the clean population that descriptors, structural filters, synthesis, and docking see.

Behavior

Controlled by config_mol_prep.yml.

Normalization (steps.*):

Parse SMILES to RDKit molecules (to_mol)
Optional fix / sanitize
Remove salts and solvents; keep the largest fragment
Standardize: disconnect metals, normalize, reionize
Preserve formal charge by default (standardize_mol.uncharge: false)
Preserve stereochemistry by default (remove_stereochemistry: false)
Write standardized SMILES

Hard filters (filters.*):

Key	Default	Effect
`allowed_atoms`	`C,N,O,S,F,Cl,Br,I,P,H,Si`	Reject unsupported elements
`reject_radicals`	`true`	Reject radical electrons
`reject_isotopes`	`false`	Keep isotopes by default
`require_single_fragment`	`true`	Reject multi-fragment leftovers

set_allowed_atoms_from_targets: true replaces filters.allowed_atoms with the union of atoms seen in target_mols_path during target alignment only. Outside alignment it has no effect.

output.write_duplicates_removed: true writes duplicates_removed.csv in normal mode. In large-dataset mode the duplicates file is always written; this flag is ignored there.

Output

Under stages/01_mol_prep/:

filtered_molecules.csv — survivors
failed_molecules.csv — rejects
duplicates_removed.csv — when duplicate writing is enabled

If MolPrep finishes with zero molecules, the pipeline exits early.

Usage


uv run hedgehog
uv run hedgehog --stage mol_prep
uv run hedge --stage mol_prep