mantispy.ds.synthetic_plate

mantispy.ds.synthetic_plate(n_plates=1, n_wells=384, n_cells=200, n_features=60, channels=('DNA', 'ER', 'RNA', 'AGP', 'Mito'), n_perturbations=8, effect_size=1.0, n_images_per_well=1, n_batches=1, plate_effect=0.0, row_gradient=0.0, col_gradient=0.0, batch_effect=0.0, batch_rotation=0.0, confounder_effect=0.0, n_bad_images=0, n_correlated_pairs=0, n_constant_features=0, nan_fraction=0.0, seed=0)

Generate a synthetic plate with known ground truth.

Every injected effect is recorded under the top-level uns["truth"] (not a schema key, and not carried through aggregation) so tests and tutorials can check that a method recovers it.

Parameters:

n_plates int (default: 1)

Number of plates.

n_wells int (default: 384)

Wells per plate.

n_cells int (default: 200)

Cells per well, or the Poisson mean of that count when confounder_effect is set.

n_features int (default: 60)

Number of features, before correlated copies are added.

channels Sequence[str] (default: ('DNA', 'ER', 'RNA', 'AGP', 'Mito'))

Channel vocabulary used to build feature names. Any names work.

n_perturbations int (default: 8)

Number of treatments; a DMSO negative control is always added.

effect_size float (default: 1.0)

Shift applied to the features affected by each perturbation.

n_images_per_well int (default: 1)

Fields of view per well, which sets Metadata_ImageNumber.

n_batches int (default: 1)

Plates are assigned round-robin to this many batches.

plate_effect float (default: 0.0)

Shift added to every feature, multiplied by the plate’s index.

row_gradient float (default: 0.0)

Shift added across the plate rows, for plate-position correction.

col_gradient float (default: 0.0)

Shift added across the plate columns, for plate-position correction.

batch_effect float (default: 0.0)

Standard deviation of a random per-batch offset added to every feature.

batch_rotation float (default: 0.0)

Strength of a random per-batch rotation of the first min(5, n_features) features.

confounder_effect float (default: 0.0)

Makes a fifth of the features scale with the well’s cell count.

n_bad_images int (default: 0)

Images given a degraded image-quality profile and noisier cells.

n_correlated_pairs int (default: 0)

Near-copies of existing features to add, for feature selection to remove.

n_constant_features int (default: 0)

Features set to a constant, for feature selection to remove.

nan_fraction float (default: 0.0)

Fraction of entries set to NaN.

seed int (default: 0)

Seed for reproducibility.

Return type:

AnnData

Returns:

An AnnData at cell resolution, carrying every injected effect under uns["truth"], the channel vocabulary under channels and the per-image quality metrics under image_table.

Raises:

ValueError – n_features needs more distinct names than channels can spell, or n_wells exceeds the largest standard plate format.