omicverse.metabol.roc_feature

Contents

omicverse.metabol.roc_feature#

omicverse.metabol.roc_feature(adata, *, group_col, pos_group=None, neg_group=None, layer=None, ci=False, n_bootstrap=1000, seed=0)[source]#

Per-feature AUC for a binary class.

Parameters:

  • group_col (str) – Column in adata.obs with the class labels.

  • pos_group (Optional[str] (default: None)) – Label strings. If None the first two unique values in group_col are used (neg_group first, pos_group second — i.e. alphabetical/appearance order).

  • neg_group (Optional[str] (default: None)) – Label strings. If None the first two unique values in group_col are used (neg_group first, pos_group second — i.e. alphabetical/appearance order).

  • layer (Optional[str] (default: None)) – AnnData layer name (default None → adata.X).

  • ci (bool (default: False)) – If True compute bootstrap 95% CI per feature. Costs n_bootstrap × n_features AUC evaluations; default False.

  • n_bootstrap (int (default: 1000)) – Number of bootstrap resamples per feature. Default 1000.

  • seed (int (default: 0)) – Bootstrap RNG seed.

Returns:

Indexed by feature, sorted by AUC descending, with columns auc (and ci_low, ci_high if ci=True).

Return type:

pd.DataFrame

Contents