Sample distance

sample_distance computes the pairwise distance matrix across samples using the chosen metric. Call it once per metric; each call writes its own heatmap and CSV. Standard vector metrics (cosine, correlation, euclidean, ...) operate on the single sample embedding uns['X_DR_sample']; distribution metrics (EMD, chi_square, jensen_shannon) operate on cell-type proportions and require the cell-level AnnData via cell_adata.

Call

from sampledisco.sample_distance.sample_distance import sample_distance

for method in ["cosine", "correlation"]:
    sample_distance(
        adata=adata,
        output_dir="sampledisco_demo_output/rna",
        method=method,
        data_type="RNA",
        grouping_columns=["sev.level"],
    )

Here adata is the cell-level AnnData carrying the sample embedding in adata.uns['X_DR_sample'] (written in place by compute_sample_embedding). For the distribution metrics, also pass cell_adata=adata:

sample_distance(
    adata=adata,
    output_dir="sampledisco_demo_output/rna",
    method="EMD",
    data_type="RNA",
    cell_adata=adata,
    sample_column="sample",
    cell_type_column="cell_type",
    grouping_columns=["sev.level"],
)

Output

Writes → sampledisco_demo_output/rna/{method}/:

• sample_DR_coordinates.csv and distance_matrix_sample_DR.csv
• sample_distance_sample_DR_heatmap.pdf
• distance_statistics_summary_{method}.csv
• Group-summary results via distanceCheck when grouping_columns is set.

For EMD, outputs go under sampledisco_demo_output/rna/EMD_distance/ (distance_matrix_EMD.csv, cell_type_proportions.csv, cell_type_centroids.csv, distance_matrix_EMD_heatmap.pdf).

Result

Sample-embedding distances under the cosine metric. Samples with similar phenotype cluster in the heatmap.

Sample-embedding distances under the correlation metric.

See the API page for the full parameter list.