MolecularDiffusion.modules.tasks.pharmacophore

Pharmacophore Diffusion Task.

Multi-modal diffusion task for generating molecules with 4 modalities: - x1: Molecular structure (atoms + bonds) - x2: Shape (point cloud) - x3: Electrostatics (point cloud) - x4: Pharmacophores (points + directions)

Attributes

logger

Classes

PharmacophoreGenerative

Generative diffusion task for pharmacophore model.

Module Contents

class MolecularDiffusion.modules.tasks.pharmacophore.PharmacophoreGenerative(model=None, modality_weights: Dict[str, float] | None = None, loss_type: str = 'mse', model_params: Dict | None = None, **kwargs)

Bases: MolecularDiffusion.modules.tasks.task.Task, MolecularDiffusion.core.Configurable

Generative diffusion task for pharmacophore model.

Handles training of the 4-modal pharmacophore diffusion model with per-modality loss weighting.

build()

Build the task (no-op, model already instantiated).

evaluate(pred, target)

Return empty metrics for generative models (evaluation not supported).

forward(batch)

Forward pass through the model.

Parameters:

batch – HeteroData batch from pharmacophore_collate_fn

Returns:

Tuple of (total_loss, metrics_dict)

predict(batch: Dict[str, Any], all_loss=None, metric=None)

Return empty predictions for generative models (evaluation not supported).

preprocess(train_set=None, valid_set=None, test_set=None)

Preprocess datasets.

Parameters:

• train_set – Training dataset

• valid_set – Validation dataset

• test_set – Test dataset

target(batch: Dict[str, Any])

Return empty targets for generative models (evaluation not supported).

loss_type = 'mse'

modality_weights = None

model = None

task_type = 'diffusion_pharmacophore'

MolecularDiffusion.modules.tasks.pharmacophore.logger