scviva.tools.vision.compute_signature_scores

scviva.tools.vision.compute_signature_scores(adata, norm_data_key, signature_varm_key, sig_norm_method='znorm_columns', random_state=0)

Compute Geary’s C spatial/graph autocorrelation for all signatures.

Rank-transforms the per-cell scores in adata.obsm["vision_signatures"] and computes Geary’s C against the cell-cell neighbor graph in adata.obsp["weights"]. A null distribution is built from randomly generated background signatures (generate_permutations_null()), matched on gene-set size and up/down balance via k-means clustering, and used to derive an empirical p-value per signature (compute_sig_pvals()).

Mirrors R VISION’s signature autocorrelation step.

Parameters:

• adata (AnnData) – Annotated data object. Must contain adata.obsm["vision_signatures"] (see compute_signatures_anndata()) and adata.obsp["weights"].

• norm_data_key ({"use_raw"} or str or None) – Expression key used for the original signature scoring; also used to score the random background signatures. "use_raw" reads from adata.raw.X.

• signature_varm_key (str) – Key in adata.varm for the (n_genes x n_sigs) signature weight matrix used to derive the random background signature sizes/balance.

• sig_norm_method (str, optional) – Normalisation method used to score the random background signatures. Must match the sig_norm_method used to compute the real scores in adata.obsm["vision_signatures"] – scoring the null background under a different normalisation than the real statistic invalidates the permutation test. Default "znorm_columns".

• random_state (int, optional) – Seed forwarded to generate_permutations_null() for the random background signature generation, by default 0.

Returns:

Columns c_prime (1 - Geary’s C), pvals (empirical p-value from the matched random background), and fdr (BH-corrected).

Return type:

pd.DataFrame indexed by signature name