scib.metrics.pcr

scib.metrics.pcr(adata, covariate, embed=None, n_comps=50, recompute_pca=True, verbose=False)

Principal component regression for anndata object

Wraps pc_regression() while checking whether to:

  • compute PCA on embedding or expression data (set embed to name of embedding matrix e.g. embed='X_emb')

  • use existing PCA (only if PCA entry exists)

  • recompute PCA on expression matrix (default)

Parameters:

  • adata – Anndata object

  • covariate – Key for adata.obs column to regress against

  • embed – Embedding to use for principal component analysis. If None, use the full expression matrix (adata.X), otherwise use the embedding provided in adata.obsm[embed].

  • n_comps – Number of PCs, if PCA is recomputed. The PCA will be recomputed if neither PCA loadings nor the principle components can be found.

  • recompute_pca – whether to recompute a PCA on the

Returns:

Variance contribution of regression

The function can be computed on full corrected feature spaces and latent embeddings. No preprocessing is needed, as the function can perform PCA if recompute_pca=True. Alternatively, you can also provide precomputed PCA, if the principle components are saved under .obsm["X_pca"] and the PC loadings are saved in .uns["pca"]["variance"].

Examples

# full feature output
scib.me.pcr(adata, covariate="batch", recompute_pca=True)

# embedding output
scib.me.pcr(adata, covariate="batch", embed="X_emb")