RAISIN cluster DGE¶

Two-step cluster-level differential expression: raisinfit fits the RAISIN hierarchical GLM (mean + sample-level + cell-level variance components) given a sample-to-cluster mapping, then run_pairwise_tests iterates over every group pair and emits volcano plots, results CSVs, and cross-cluster summary figures.

Call¶

from sampledisco.sample_clustering.RAISIN import raisinfit
from sampledisco.sample_clustering.RAISIN_TEST import run_pairwise_tests

fit = raisinfit(
    adata=adata_cell,
    sample_col="sample",
    sample_to_clade=expr_clusters,     # from cluster(...)
    testtype="unpaired",
    group_col=None,
    batch_col=None,
    intercept=True,
    filtergene=False,
    n_jobs=8,
    verbose=True,
)

run_pairwise_tests(
    fit=fit,
    output_dir="sampledisco_demo_output/rna/raisin_results_expression",
    fdrmethod="fdr_bh",
    n_permutations=100,
    fdr_threshold=0.05,
    top_n_genes=50,
    make_summary_plots=True,
)

Output¶

Writes → sampledisco_demo_output/rna/raisin_results_expression/:

One subdirectory per pair (0_vs_1/, 0_vs_2/, ...) each containing raisin_results.csv, volcano.png, and a labeled variant.
summary_plots/pseudobulk_heatmap.png — top DE genes across all clusters.
summary_plots/summary_dotplot.png — per-pair DE counts.
summary_plots/all_results_combined.csv.

Result¶

Example pairwise volcano (cluster 0 vs 1) Pseudobulk heatmap across clusters Summary dotplot of pairwise DE counts

Left: one pair's volcano. Middle: cross-cluster pseudobulk heatmap of top genes. Right: DE gene count per comparison.

See the API pages for raisinfit and run_pairwise_tests.