Partition PBMC 68k (reduced)

Partition Scanpy’s bundled pbmc68k_reduced dataset with CAMP3, then compute quality metrics against Louvain clusters.

Setup and partition

import campmc as cp
import scanpy as sc
adata = sc.datasets.pbmc68k_reduced()
cp.partition(adata, method="camp3", gamma=50, random_state=0)
adata.obs["camp"].head()

index
AAAGCCTGGCTAAC-1    seed11-50-__all__
AAATTCGATGCACA-1    seed11-50-__all__
AACACGTGGTCTTT-1     seed3-50-__all__
AAGTGCACGTGCTA-1    seed13-50-__all__
ACACGAACGGAGTG-1     seed5-50-__all__
Name: camp, dtype: str

This object already has PCA embeddings, so preprocessing is optional. Labels land in obs['camp']; coreset seeds are flagged in obs['camp_is_seed'].

import campmc as cp
import scanpy as sc
adata = sc.datasets.pbmc68k_reduced()
cp.partition(adata, method="camp3", gamma=50, random_state=0)
int(adata.obs["camp_is_seed"].sum())

14

Quality metrics

import campmc as cp
import scanpy as sc
adata = sc.datasets.pbmc68k_reduced()
cp.partition(adata, method="camp3", gamma=50, random_state=0)
metrics = cp.mt.quality(adata, partition_key="camp", label_key="louvain")
metrics[["metacell_id", "size", "purity", "compactness"]].head()

	metacell_id	size	purity	compactness
0	seed0-50-__all__	65	0.984615	-0.001010
1	seed1-50-__all__	34	0.911765	-0.000305
2	seed10-50-__all__	3	0.666667	-0.000724
3	seed11-50-__all__	158	0.708861	-0.000695
4	seed12-50-__all__	42	0.809524	-0.000890

Variants

• Other assigners: method="camp1" … "camp4". CAMP4 uses SEACells and is slower on large objects.

• Multiple compressions: gammas=[25, 50, 100] writes camp_25, camp_50, …

• Raw counts: run campmc.preprocess() first to obtain obsm['X_pca'].