Tasks

Tasks map sampler indices to model inputs and supervision targets. Summaries below are taken from each class docstring.

Predict amino-acid sequences from backbone structure.

class bioverse.tasks.inverse_folding.InverseFoldingTask(resolution='residue')

Bases: Task

Predict amino-acid sequences from backbone structure.

Predict masked residue identities or properties.

class bioverse.tasks.masked_residue_prediction.MaskedResiduePredictionTask

Bases: Task

Predict masked residue identities or properties.

Predict the effect of amino-acid mutations.

class bioverse.tasks.mutation_effect_prediction.MutationEffectPredictionTask(resolution='residue')

Bases: Task

Predict the effect of amino-acid mutations.

Predict a property for pairs of molecules.

class bioverse.tasks.pairwise_property_prediction.PairwisePropertyPredictionTask(meta: str, id_field: str = 'id', level: str = 'molecule', target: str | None = None, resolution: str = 'residue')

Bases: Task

Predict a property for pairs of molecules.

Predict a scalar or vector property at a chosen hierarchy level.

class bioverse.tasks.property_prediction.PropertyPredictionTask(property='label', level='molecule', resolution='atom')

Bases: Task

Predict a scalar or vector property at a chosen hierarchy level.

Loads a molecule (or sub-molecule level) from the indexed batch and extracts {level}_{property} as the supervision target. Set resolution to control which atoms/residues are exposed as features.

Parameters:

• property (str or tuple) – Target column name, or (name, index) to select one element of a vector-valued property.

• level (str) – Hierarchy level of the target ("molecule", "residue", etc.).

• resolution (str) – Feature resolution passed to the returned batch ("atom" or "residue").

Score ligand-target pairs for virtual screening.

class bioverse.tasks.virtual_screen.VirtualScreenTask

Bases: Task

Score ligand-target pairs for virtual screening.