pyucell.smooth_knn_scores

pyucell.smooth_knn_scores#

pyucell.smooth_knn_scores(adata, obs_columns, k=10, use_rep='X_pca', decay=0.1, up_only=False, graph_key=None, suffix='_kNN', device=None)#

Smooth per-cell signature scores by weight-averaging over k nearest neighbors.

Parameters:

  • adata (AnnData) – AnnData with .obs columns containing signature scores.

  • obs_columns (list of str) – Names of columns in adata.obs with the scores to smooth.

  • k (int) – Number of neighbors (ignored if graph_key provided).

  • use_rep (str) – Representation to compute neighbors (e.g., ‘X_pca’, ‘X’).

  • decay (float) – Decay parameter (0<decay<1) for weighting nearest neighbors: weight = (1 - decay)^i for i-th neighbor.

  • up_only (bool) – If True, ensures smoothed scores are at least as high as the original.

  • graph_key (str or None) – If provided, the name of a precomputed graph in adata.obsp to use (e.g. ‘connectivities’ or ‘my_graph’). If None, compute neighbors with scanpy.

  • suffix (str) – Optional suffix for smoothed scores. By default ‘_kNN’ is appended.

  • device (str | None, optional) – None (default) → CPU sparse matmul. "cpu" / "cuda" / "mps" / "auto" → torch sparse matmul on that device. Requires the pyucell[gpu] extra.

Returns:

None. Adds smoothed scores to adata.obs.