The AnnData Object

Scanpy is Python’s leading single-cell analysis framework. This chapter covers the equivalent workflow to Seurat.

The AnnData structure

# AnnData is the core data structure
# adata.X         - count matrix (cells x genes)
# adata.obs       - cell metadata (DataFrame)
# adata.var       - gene metadata (DataFrame)
# adata.obsm      - embeddings (PCA, UMAP)
# adata.layers    - alternative matrices (raw, normalized)
# adata.uns       - unstructured data

Loading a 10x matrix

import scanpy as sc

sc.settings.verbosity = 3

# The PBMC 3k matrix ships in the section container at /opt/data/pbmc3k.
adata = sc.read_10x_mtx("/opt/data/pbmc3k", var_names="gene_symbols",
                        make_unique=True)

# Store the raw counts before anything modifies adata.X.
adata.layers["counts"] = adata.X.copy()

print(adata)

AnnData object with n_obs × n_vars = 2700 × 32738
    var: 'gene_ids'
    layers: None (.X), 'counts'

print(adata.var.head())
print(adata.X[:4, :4].toarray())

                     gene_ids
MIR1302-10    ENSG00000243485
FAM138A       ENSG00000237613
OR4F5         ENSG00000186092
RP11-34P13.7  ENSG00000238009
RP11-34P13.8  ENSG00000239945
[[0. 0. 0. 0.]
 [0. 0. 0. 0.]
 [0. 0. 0. 0.]
 [0. 0. 0. 0.]]

Sparsity

Typical scRNA-seq data is 90-95% zeros:

Low expression genes not detected
Technical dropout
True biological zeros

zero_fraction = 1.0 - adata.X.nnz / (adata.n_obs * adata.n_vars)
print(f"fraction of zero entries: {zero_fraction:.4f}")

fraction of zero entries: 0.9741

Saving and loading

adata.write_h5ad("outputs/anndata-object_pbmc3k.h5ad")

adata_copy = sc.read_h5ad("outputs/anndata-object_pbmc3k.h5ad")
print(adata_copy)

AnnData object with n_obs × n_vars = 2700 × 32738
    var: 'gene_ids'
    layers: 'counts', None (.X)