cellink.tl.external.get_disease_relevant_cells

cellink.tl.external.get_disease_relevant_cells(weights_df, adata, top_percentile=85.0, fdr_threshold=0.1, cell_type_col=None)

Identify disease-relevant cells from scPRS model weights.

Adaptation of util/analysis.py (szhang1112/scPRS):

1. For each repeat, compute the top_percentile-th percentile of all cell weights as a scalar reference value.

2. For each cell, one-sided t-test comparing its weight distribution across repeats against the per-repeat percentile values.

3. Benjamini-Hochberg FDR correction (method='n' as in the paper).

Parameters:

• weights_df (pd.DataFrame) – Cell weight matrix (n_repeats c n_cells) from run_scprs().

• adata (AnnData) – Original scATAC-seq AnnData whose obs_names correspond to cell indices in weights_df.columns.

• top_percentile (float, default=85.0) – Percentile used to define the reference distribution per repeat. The paper uses the 85th percentile (top 15%).

• fdr_threshold (float, default=0.1) – Adjusted p-value threshold (BY correction, as in analysis.py).

• cell_type_col (str, optional) – adata.obs column with cell-type labels for enrichment analysis.

Return type:

DataFrame | tuple[DataFrame, DataFrame]

Returns:

pd.DataFrame or tuple If cell_type_col is None: DataFrame with per-cell results. If cell_type_col is provided: tuple of (cell_results, ct_results).