MolecularDiffusion.modules.models.pharmacophore_dynamics¶

Pharmacophore Model Dynamics - Wrapper for ShEPhERD model.

Builds input_dict from HeteroData batch exactly matching shepherd/src/shepherd/lightning_module.py get_training_input_dict(), then calls shepherd_arch.Model.forward().

Attributes¶

logger

Classes¶

`PharmacophoreEnVariationalDiff`	Wrapper that makes ShEPhERD compatible with MolecularDiffusion's
`PharmacophoreModelWrapper`	Wrapper around ShEPhERD Model.

Module Contents¶

class MolecularDiffusion.modules.models.pharmacophore_dynamics.PharmacophoreEnVariationalDiff(model_params: Dict | None = None, device: torch.device | None = None)¶

Bases: torch.nn.Module

Wrapper that makes ShEPhERD compatible with MolecularDiffusion’s EnVariationalDiffusion interface.

forward(batch, device=None)¶

device = None¶

dynamics¶

class MolecularDiffusion.modules.models.pharmacophore_dynamics.PharmacophoreModelWrapper(shepherd_model, params=None)¶

Bases: torch.nn.Module

Wrapper around ShEPhERD Model.

Converts a batched HeteroData into the input_dict format that shepherd_arch.Model.forward() expects, runs the forward pass, and returns (input_dict, output_dict).

forward(batch, device=None) → Tuple[Dict[str, Any], Dict[str, Any]]¶

Build input_dict from HeteroData batch and run model.

Parameters:

batch – HeteroData batch from pharmacophore_collate_fn
device – target device

Returns:

(input_dict, output_dict)

model¶

params¶

MolecularDiffusion.modules.models.pharmacophore_dynamics.logger¶