campa.tl.extract_features

extract_features(params)[source]

Extract features from clustered dataset using FeaturesExtractor.

Creates features anndata.AnnData object.

Parameters determine what features are extracted from a given clustering. The following keys in params are expected:

experiment_dir: path to experiment directory relative to campa_config.EXPERIMENT_DIR.
cluster_name: name of clustering to use.
cluster_dir: dir of subsampled clustering to load annotation. Relative to experiment_dir. Default is taking first of experiment_dir/aggregated/sub-*.
cluster_col: cluster annotation to use. Defaults to cluster_name.
data_dirs: data directories to be processed. Relative to experiment_dir/aggregated/full_data. If None, all available data_dirs will be processed.
save_name: filename to use for saving extracted features.
force: force calculation even when adata exists.
features: type of features to extract. One or more of intensity, co-occurrence, object-stats.
- Intensity: per-cluster mean and size features. Needs to be calculated first to set up the adata.
- Co-occurrence: spatial co-occurrence between pairs of clusters at different distances.
- Object stats: number and area of connected components per cluster.
co_occurrence_params: parameters for co-occurrence calculation.
- min, max, nsteps: size of distances interval.
- logspace: use log spacing of co-occurrence intervals.
- num_processes: number of processes to use to compute co-occurrence scores.
object_stats_params: parameter dict for object-stats calculation.
- features: features to extract in mode object-stats. Possible features: area, circularity, elongation, extent.
- channels: intensity channels to extract mean per cluster from.