campa.tl.Cluster

class Cluster(config, cluster_mpp=None, save_config=False)

Cluster data.

Contains functions to create a (subsampled) MPPData for clustering, cluster it, and to project the clustering to other MPPDatas.

Cluster is initialised from a cluster config dictionary with the following keys:

• data_config: name of the data config to use, should be registered in campa.ini

• data_dirs: where to read data from (relative to DATA_DIR defined in data config)

• process_like_dataset: name of dataset that gives parameters for processing (except subsampling/subsetting)

• subsample: (bool) subsampling of pixels

• subsample_kwargs: kwargs for MPPData.subsample() defining the fraction of pixels to be sampled

• subset: (bool) subset to objects with certain metadata.

• subset_kwargs: kwargs to MPPData.subset() defining which object to subset to

• seed: random seed to make subsampling reproducible

• cluster_data_dir: name of the dir containing the mpp_data that is clustered. Relative to EXPERIMENT_DIR

• cluster_name: name of the cluster assignment file

• cluster_rep: representation that should be clustered (name of existing file, should be predicted with Predictor.get_representation()).

• cluster_method: leiden or kmeans (kmeans not tested).

• leiden_resolution: resolution parameter for leiden clustering.

• kmeans_n: number of clusters for kmeans.

• umap: (bool) predict UMAP of cluster_rep.

Parameters

• config (MutableMapping[str, Any]) – Cluster config.

• cluster_mpp (Optional[MPPData]) – Data to cluster.

• save_config (bool) – Save cluster config in {config['cluster_data_dir']}/cluster_params.json.

Attributes

cluster_annotation	Cluster annotation pd.DataFrame, read from {cluster_name}_annotation.csv.
cluster_mpp	MPPData that is used for clustering.
config	Cluster config.

Methods

add_cluster_annotation(annotation, to_col[, ...])	Add annotation and colormap to clustering.
add_cluster_colors(colors[, from_col])	Add colours to clustering or to annotation.
add_umap()	If umap does not yet exist, but should be calculated, calculates umap.
create_cluster_mpp()	Use cluster parameters to create and save Cluster.cluster_mpp to use for clustering.
create_clustering()	Cluster Cluster.cluster_mpp using cluster_method defined in Cluster.config.
from_cluster_data_dir(data_dir)	Initialise from existing cluster_data_dir.
from_exp(exp[, cluster_config, data_dir])	Initialise from experiment for clustering of entire data that went into creating training data.
from_exp_split(exp)	Initialise from experiment for clustering of val/test split.
get_hpa_localisation([cluster_name, thresh, ...])	Query subcellular localisation for each cluster from Human Protein Atlas (https://www.proteinatlas.org).
get_nndescent_index([recreate])	Calculate and return pynndescent index of existing clustering for fast prediction of new data.
predict_cluster_imgs(exp)	Predict cluster images from experiment.
predict_cluster_rep(exp)	Use experiment to predict the necessary cluster representation.
project_clustering(mpp_data[, save_dir, ...])	Project already computed clustering from Cluster.cluster_mpp to mpp_data.
set_cluster_name(cluster_name)	Change the cluster name and reloads cluster_mpp, and cluster_annotation.