schrodinger.application.transforms.screening module

class schrodinger.application.transforms.screening.FixedSizeSamplingConfig(*, sample_size: int, distinct: bool = False, seed: int | None = None)

Bases: BaseModel

Configuration for fixed sampling before an expensive screening transform.

Parameters:

  • sample_size – fixed number of inputs to sample

  • distinct – deduplicate inputs before sampling

  • seed – random seed for reproducible approximate sampling

sample_size: int
distinct: bool
seed: int | None
model_config: ClassVar[ConfigDict] = {'frozen': True}

Configuration for the model, should be a dictionary conforming to [ConfigDict][pydantic.config.ConfigDict].

class schrodinger.application.transforms.screening.ScreenMoleculesConfig(*, property_space: schrodinger.application.transforms.filters.PropertySpaceFilterConfig | None = None, complexity: schrodinger.application.transforms.filters.ComplexityFilterConfig | None = None, substructure_filter_path: ~typing.Optional[~pathlib.Path] = None, fep_amenability: schrodinger.application.transforms.filters.FepAmenableConfig | None = None, similarity: schrodinger.application.transforms.filters.RDKitPathTanimotoSimilarityFilterConfig | None = None, pre_ligprep_sampling: schrodinger.application.transforms.screening.FixedSizeSamplingConfig | schrodinger.seam.transforms.samplers.ThrottleConfig | None = None, ligprep: ~schrodinger.application.transforms.ligprep.core.LigPrepConfig = <factory>, ligprep_property_ranges: dict[typing.Annotated[str, AfterValidator(func=<function _number_property_validator at 0x7d07c41e58a0>)], typing.Annotated[tuple[int | float, int | float], AfterValidator(func=<function _range_validator at 0x7d07c41e5760>)]] = <factory>, charged_nitrogen_pka: schrodinger.application.transforms.filters.ChargedNHpKaFilterConfig | None = None, pre_docking_sampling: schrodinger.application.transforms.screening.FixedSizeSamplingConfig | schrodinger.seam.transforms.samplers.ThrottleConfig | None = None, docking: ~schrodinger.application.transforms.dock.DockingConfig, hypothesis: schrodinger.application.transforms.filters.HypothesisFilterConfig | None = None)

Bases: BaseModel

Configuration for ScreenMolecules.

Optional filter configurations enable their corresponding transforms. The supplied substructure filter file is applied both before LigPrep and before docking. ligprep_property_ranges filters numeric properties available on the LigPrep output structures.

Distinct pre-LigPrep sampling is supported when the candidates are molecules. Similarity filtering first converts them to structures, which cannot safely be passed to Beam’s generic distinct operation. Pre-docking candidates have already been deduplicated by molecular SMILES, so distinct pre-docking sampling is unnecessary and unsupported.

property_space: PropertySpaceFilterConfig | None
complexity: ComplexityFilterConfig | None
substructure_filter_path: Optional[Path]
fep_amenability: FepAmenableConfig | None
similarity: RDKitPathTanimotoSimilarityFilterConfig | None
pre_ligprep_sampling: FixedSizeSamplingConfig | ThrottleConfig | None
ligprep: LigPrepConfig
ligprep_property_ranges: dict[typing.Annotated[str, AfterValidator(func=<function _number_property_validator at 0x7d07c41e58a0>)], typing.Annotated[tuple[int | float, int | float], AfterValidator(func=<function _range_validator at 0x7d07c41e5760>)]]
charged_nitrogen_pka: ChargedNHpKaFilterConfig | None
pre_docking_sampling: FixedSizeSamplingConfig | ThrottleConfig | None
docking: DockingConfig
hypothesis: HypothesisFilterConfig | None
model_config: ClassVar[ConfigDict] = {'frozen': True}

Configuration for the model, should be a dictionary conforming to [ConfigDict][pydantic.config.ConfigDict].

class schrodinger.application.transforms.screening.ScreenMoleculesOutputs(ligprep_candidates: PCollection[rdkit.Chem.rdchem.Mol | Structure], after_ligprep: PCollection[Structure], docking_inputs: PCollection[Structure], after_docking: PCollection[Structure], main: PCollection[Structure])

Bases: object

Intermediate and final PCollections available for reporting and diagnostic side branches.

Additional transforms may be attached to the intermediate collections without coupling ScreenMolecules to a particular metrics backend or workflow-specific diagnostic behavior.

ligprep_candidates

Molecules or structures that passed all pre-LigPrep filters, before pre-LigPrep sampling or reshuffling. This collection contains molecules unless similarity filtering is enabled, in which case it contains structures.

after_ligprep

Structures after LigPrep, structure deduplication, and LigPrep-property filtering, but before the charged-nitrogen and second substructure filters.

docking_inputs

Structures after all pre-docking filters and pre-docking sampling or reshuffling. These are the structures supplied to docking.

after_docking

Structures produced by docking, before the optional hypothesis filter.

main

Final structures after the optional hypothesis filter.

ligprep_candidates: PCollection[rdkit.Chem.rdchem.Mol | Structure]
after_ligprep: PCollection[Structure]
docking_inputs: PCollection[Structure]
after_docking: PCollection[Structure]
main: PCollection[Structure]
__init__(ligprep_candidates: PCollection[rdkit.Chem.rdchem.Mol | Structure], after_ligprep: PCollection[Structure], docking_inputs: PCollection[Structure], after_docking: PCollection[Structure], main: PCollection[Structure]) → None
class schrodinger.application.transforms.screening.ScreenMolecules(*, label: str | None = None, funnel_input_label: str = 'Inputs', substructure_filter: SubstructureFilter | None = None, **kwargs)

Bases: PTransformWithConfig

Apply a configurable molecular screening cascade.

The cascade filters RDKit molecules, converts them to structures, prepares them with LigPrep, filters the prepared structures, docks them, and optionally applies a Phase hypothesis filter.

Parameters:

  • label – optional Beam transform label; this is a constructor-only argument and is not part of ScreenMoleculesConfig

  • funnel_input_label – label used for the funnel’s input collection; this is also a constructor-only argument

  • substructure_filter – optional prepared substructure filter to apply before LigPrep and before docking; this is a constructor-only argument and cannot be combined with substructure_filter_path

Example usage:

import apache_beam as beam

from schrodinger.application.transforms.screening import (
    ScreenMolecules)
from schrodinger.application.transforms.screening import (
    ScreenMoleculesConfig)
from schrodinger.application.transforms.dock import DockingConfig

config = ScreenMoleculesConfig(
    docking=DockingConfig(grid_path='receptor.zip'))
screened = (
    mols
    | ScreenMolecules.from_config(config)
    | 'Reshuffle screened molecules' >> beam.Reshuffle()
)

A caller that has already prepared a substructure filter can inject it without writing another filter file:

outputs = mols | ScreenMolecules.from_config(
    config, substructure_filter=prepared_filter)

Intermediate PCollections are available for reporting and diagnostic transforms:

outputs = mols | ScreenMolecules.from_config(config)
screened = outputs.main
after_ligprep_count = (
    outputs.after_ligprep
    | 'Count after LigPrep' >> beam.combiners.Count.Globally()
)
config_class

alias of ScreenMoleculesConfig

LIGPREP_CANDIDATES_COUNT_SOURCE = 'ligprep_candidates'
AFTER_LIGPREP_COUNT_SOURCE = 'after_ligprep'
AFTER_DOCKING_COUNT_SOURCE = 'after_docking'
MAIN_COUNT_SOURCE = 'main'
__init__(*, label: str | None = None, funnel_input_label: str = 'Inputs', substructure_filter: SubstructureFilter | None = None, **kwargs)

Initialize with constructor-only runtime options.

display_data()

Return configuration and filter-funnel metadata.

expand(mols)

Expand the cascade and return its outputs.

schrodinger.application.transforms.screening.screening_output_metrics(group_path: tuple[str, ...], transform_label: str, *, before_ligprep_key: str, after_ligprep_key: str, after_docking_key: str, output_key: str) → OutputMetricsDisplayData

Declare output metrics for the standard screening outputs.

Parameters:

  • group_path – common prefix of each output metric path

  • transform_label – label of ScreenMolecules relative to the transform declaring the metrics

  • before_ligprep_key – key for candidates before LigPrep sampling

  • after_ligprep_key – key for prepared, deduplicated structures

  • after_docking_key – key for structures produced by docking

  • output_key – key for the final filtered structures