sc_reconstruction.metrics.metric_deg#

sc_reconstruction.metrics.metric_deg(adata_true, adata_pred, *, ref_true, ref_pred, method='wilcoxon', dice_k=(100,), min_cells=20, set_neg_to_zero=False, fdr_threshold=0.05)[source]#

Differential-expression recovery vs a reference (e.g. control) condition.

The Wilcoxon and t-test calls are forwarded to scanpy [Wolf et al., 2018]. The metric computes DEGs of (true vs ref_true) and (pred vs ref_pred), then reports overlap (Dice@k) and rank-correlation on the log-fold-changes. Wraps DegCalculator.compute_deg().

Parameters:
  • ref_true (anndata.AnnData) – Reference (e.g. control) condition’s true and predicted expression.

  • ref_pred (anndata.AnnData) – Reference (e.g. control) condition’s true and predicted expression.

  • method (str) – "wilcoxon" or "t-test". Forwarded to scanpy [Wolf et al., 2018].

  • dice_k (Sequence[int]) – K values at which to compute Dice overlap of top-K DEGs.

  • adata_true (anndata.AnnData)

  • adata_pred (anndata.AnnData)

  • min_cells (int)

  • set_neg_to_zero (bool)

  • fdr_threshold (float | None)

Return type:

dict[str, float]