MolecularDiffusion.utils.conformer_pool

Conditioning-molecule loading for conformer generation.

Shared by every conformer-generating model. Lifted verbatim out of modules/tasks/diffusion_loqi.py, which is where it was first written and which still re-exports it for backward compatibility – a loader used by three models does not belong in one model’s task module, and runmodes/generate/tasks_conformer.py must be able to reach it without importing a heavy model task.

load_conditioning_pool is the single documented way to turn a user’s sample_input (a .sdf, a .smi/.txt, a graph3d ASE .db, or an inline list of SMILES) into the platform’s graph3d per-item Data.

Attributes

logger

Functions

load_conditioning_pool(→ list[torch_geometric.data.Data])

Load conditioning molecules.

Module Contents

MolecularDiffusion.utils.conformer_pool.load_conditioning_pool(sample_input: str | collections.abc.Sequence[str], atom_vocab: list[str], limit: int | None = None) → list[torch_geometric.data.Data]

Load conditioning molecules.

sample_input is either a path – .sdf, .smi/.txt, or an ASE .db written by the graph3d converter – or a plain list of SMILES strings given inline in the generate config.

Both SMILES forms go through _pool_from_smiles().

MolecularDiffusion.utils.conformer_pool.logger