campmc.mt.recovery_scores

campmc.mt.recovery_scores(adata, *, partition_key, label_key, rare_threshold=0.01)

Score how well a partition recovers rare cell types.

Assigns each metacell the majority label among its members, then computes balanced accuracy over rare types plus ARI/NMI restricted to rare cells. Rare types are those with frequency at most rare_threshold.

Parameters:

adata AnnData

Annotated data with partition and cell-type columns in obs.

partition_key str

Column in adata.obs with metacell assignments.

label_key str

Column in adata.obs with ground-truth (or reference) labels.

rare_threshold float (default: 0.01)

Maximum fraction of cells for a type to be considered rare.

Return type:

DataFrame

Returns:

DataFrame Single-row table with balanced_accuracy, ari, nmi, and n_rare_types.

Examples

import campmc as cp
import scanpy as sc
adata = sc.datasets.pbmc68k_reduced()
cp.partition(adata, method="camp3", gamma=50, random_state=0)
cp.mt.recovery_scores(
    adata,
    partition_key="camp",
    label_key="louvain",
    rare_threshold=0.05,
)

	balanced_accuracy	ari	nmi	n_rare_types
0	0.5	0.974249	0.967538	4