symphonypy.tl.per_cluster_confidence

symphonypy.tl.per_cluster_confidence(adata_query: AnnData, adata_ref: AnnData, cluster_key: str, u: float = 2, lamb: float = 0, transferred_primary_basis: str = 'X_pca_reference', obs: str | None = 'symphony_per_cluster_dist', uns: str | None = 'symphony_per_cluster_dist')

Calculates the Mahalanobis distance from user-defined query clusters to their nearest reference centroid after initial projection into reference PCA space. All query cells in a cluster get the same score. Higher distance indicates less confidence. Due to the instability of estimating covariance with small numbers of cells, we do not assign a score to clusters smaller than u * d, where d is the dimensionality of the embedding and u is specified.

Parameters:

  • adata_query (AnnData) – query adata object mapped to adata_ref with Symphony

  • adata_ref (AnnData) – reference adata object (with Harmony object in adata_ref.uns)

  • cluster_key (str) – which keys from adata_query.obs to use as a cluster label (if list, adata_query will be grouped by them)

  • u (float, optional) – at least u * d cells are to be assigned to a cluster, where d is a dimensionality of representation, defaults to 2

  • lamb (float, optional) – ridge regression like coef for covariance matrix inversion numerical stability, defaults to 0

  • transferred_primary_basis (str, optional) – adata_query.obsm[transferred_primary_basis] should contain pre-Symphony reference PC query representation, defaults to “X_pca_reference”

  • obs (str | None, optional) – If not None, resulted dists would be written to adata_query.obs[obs] for each cell (just the same value for each cluster), defaults to “symphony_per_cluster_dist”

  • uns (str | None, optional) – If not None, resulted dists would be written to adata_query.uns[uns] for each cluster, defaults to “symphony_per_cluster_dist”