RNA API¶

The RNA branch of SampleDisco takes a cell-level .h5ad with raw counts and turns it into a single sample-level embedding (uns['X_DR_sample']). Two RNA-specific functions live in this section; sample embedding is shared across modalities and lives under Shared.

Function	Purpose
`preprocess`	QC filter, normalize, log-transform, select HVGs, PCA, 2-pass Harmony. Writes one `adata_preprocessed.h5ad` carrying `obsm['Z_clust']` (sample-removed) and `obsm['Z_rmd']` (sample-preserved).
`cell_types`	Leiden clustering on the sample-removed `Z_clust` embedding, with optional adaptive resolution tuning to hit a target cluster count and optional UMAP. Writes labels into `.obs["cell_type"]`.

Shared with ATAC¶

compute_sample_embedding — composition blocks on Z_clust + RMD displacement block on Z_rmd, reduced to the single key uns['X_DR_sample'].
run_autotune — parameter selection: optimizes the RMD-weight α (composition vs displacement), enabled via the rna_autotune_enable flag in the config-driven wrapper.

Typical order¶

preprocess
   └─> cell_types
          └─> compute_sample_embedding
                 └─> downstream analyses (see Downstream section)

For a runnable walkthrough, see the RNA tutorial.