Downstream API¶

Everything in this section runs off the single sample embedding uns['X_DR_sample'] — the final sample-level DataFrame (units × PCs) written in place by compute_sample_embedding. The functions are identical across RNA, ATAC, and multi-omics, and most share a common wrapper invocation inside downstream_analysis().

Sample distance¶

Function	Purpose
`sample_distance`	Unified entry point for vector metrics (cosine, correlation, euclidean, ...), EMD, chi-square, Jensen–Shannon.

Dimension association¶

Function	Purpose
`run_dimension_association_analysis`	Per-PC variance-explained decomposition against every sample-level metadata variable, with permutation p-values — confounder / leading-covariate diagnostic.

Trajectory analysis¶

Function	Purpose	When to use
`CCA_Call`	Supervised pseudotime via canonical correlation with a phenotype.	You have a phenotype gradient in `.obs`.
`cca_pvalue_test`	Permutation p-value for a CCA correlation.	Want to test significance of a CCA result.
`TSCAN`	Unsupervised pseudotime via GMM + MST.	No phenotype; let structure emerge from the embedding.
`run_trajectory_gam_differential_gene_analysis`	Per-gene GAM against pseudotime, with heatmap, volcano, and curve plots.	Pseudotime (from CCA or TSCAN) in hand; find trajectory-associated genes.

Sample clustering and cluster comparisons¶

Function	Purpose	When to use
`cluster`	K-means on the sample embedding.	Want a discrete sample grouping when no phenotype is available.
`proportion_test`	eBayes moderated test on per-sample cell-type proportions across groups.	Ask whether cluster/group differences are driven by cell-type composition.
`raisinfit`	Fit the RAISIN hierarchical GLM.	Step 1 of cluster-level differential expression.
`run_pairwise_tests`	Run every pairwise contrast on a fitted RAISIN model, emit volcanos + summary plots.	Step 2 of cluster-level DE.

Visualization¶

Function	Purpose
`visualization`	Cell-type dendrogram, proportion PCA, expression UMAP — toggle which panels you want.
`visualize_multimodal_embedding`	Scatter of the sample embedding with flexible coloring; multi-omics only.

For an end-to-end walkthrough that chains these functions together, see the Downstream tutorials.