Data guide

scDLKit works directly with AnnData.

What scDLKit expects

• an AnnData object with a usable expression matrix in adata.X or a named layer

• optional label information in adata.obs

• optional batch information in adata.obs

prepare_data

Use prepare_data when you want lower-level control over preprocessing and split construction:

from scdlkit import prepare_data

prepared = prepare_data(
    adata,
    layer="X",
    label_key="louvain",
    batch_key="batch",
    use_hvg=True,
    n_top_genes=2000,
    normalize=True,
    log1p=True,
    batch_aware_split=True,
)

Scanpy-backed preprocessing

When you request normalize, log1p, or use_hvg, scDLKit uses Scanpy-backed preprocessing. Install it with:

python -m pip install "scdlkit[scanpy]"

Recommended practice

For the public tutorials, keep preprocessing simple and standard:

• use scanpy.datasets.pbmc3k_processed() for the example notebooks

• use louvain as the label field for representation and classification demos

• focus on model behavior rather than broad biological analysis