campa.data.MPPData

class MPPData(metadata, channels, data, data_config, seed=42)

Pixel-level data representation.

Backed by on-disk numpy and csv files containing intensity information per channel and metadata for each pixel. When possible, the on-disk numpy files are loaded lazily using np.memmap().

Parameters

• metadata (DataFrame) – Cell-level metadata. Needs to contain at least an data_config.OBJ_ID column, which contains cell identifiers that are used in the obj_ids data to map pixels to cells.

• channels (DataFrame) – Channel-level metadata. The first column is assumed to be the index, second column the name of the channel.

• data (Dict[str, ndarray]) – Dictionary containing pixel-level data, at least containing the required keys: x, y, obj_ids (spatial coordinates for every pixel, and assignment of pixels to objects (cells)). If mpp (per-channel pixel intensity information) is not present, is replaced with zero-value array of shape: #pixels x 1 x 1 x #channels.

• data_config (str) – Name of the data_config file registered in campa_config.data_configs.

• seed (int) – Random seed for subsampling and subsetting.

Attributes

center_mpp	#pixels x #channels array of centre pixel.
conditions	Condition information for each pixel.
has_neighbor_data	Flag indicating if neighbour information is contained in this object.
latent	Latent space for each pixel.
mpp	Multiplexed pixel profiles.
obj_ids	Object ids mapping pixels to objects (cells).
unique_obj_ids	Return unique objects (cells) contained in this object.
x	Spatial x coordinates for each pixel.
y	Spatial y coordinates for each pixel.
channels	Intensity channels.
metadata	Object (cell) level metadata (e.g.

Methods

add_conditions(cond_desc[, cond_params])	Add conditions using MPPData.metadata columns.
add_data_from_dir(data_dir[, keys, ...])	Add data to MPPData from data_dir.
add_neighborhood([size, copy])	Add a square neighbourhood around each pixels to the mpp_data.
apply_mask(mask[, copy])	Return new MPPData by applying mask to MPPData.data().
concat(objs)	Concatenate multiple MPPData objects.
copy()	Copy MPPData.
data(key)	Information contained in MPPData.
extract_csv([data, obs])	Extract information in MPPData.data() into pd.DataFrame.
from_data_dir(data_dir, data_config[, mode, ...])	Initialise MPPData from directory.
get_adata([X, obsm, obs])	Create adata from information contained in MPPData.
get_channel_ids(to_channels[, from_channels])	Map channel names to ids.
get_condition(desc[, cond_params])	Get condition based on desc.
get_img([data, channel_ids])	Calculate data image of entire MPPData.
get_object_img(obj_id[, data, channel_ids, ...])	Calculate data image of given object id.
get_object_imgs([data, channel_ids, ...])	Return images for each obj_id in current data.
normalise([background_value, percentile, ...])	Normalise MPPData.mpp values.
prepare(params)	Prepare MPP data according to given parameters (from campa.data.create_dataset()).
subsample([frac, frac_per_obj, num, ...])	Pixel-level subsampling of MPPData.
subset([frac, num, obj_ids, nona_condition, ...])	Object-level subsetting of MPPData.
subset_channels(channels)	Restrict MPPData.mpp to channels.
train_val_test_split([train_frac, val_frac])	Split along obj_ids for train/val/test split.
write(save_dir[, save_keys, mpp_params])	Write MPPData to disk.