Metrics

Metrics do not modify the object. Most return a tidy frame with metric, representation, key and value, so results from several calls stack; evaluate_integration instead draws the scib-metrics integration panel as a heatmap and returns its numeric results frame, one row per representation.

metrics.known_relationships(adata, net[, ...])	Share of annotated pairs that land in either tail of the similarity distribution [Celik et al., 2024].
metrics.evaluate_integration(adata, *[, ...])	Score one or more representations against a batch and draw the integration benchmark as a heatmap.
metrics.diagnose_testing(adata[, groupby, ...])	Check whether differential testing is calibrated on this screen.
metrics.pc_regression(adata, key[, use_rep, ...])	Variance-weighted R^2 of the principal components on key.
metrics.batch_variance_explained(adata, keys)	pc_regression() for several covariates, stacked into one frame.