Multi-omics API

The multi-omics branch integrates scRNA + scATAC via scGLUE, cleans the merged object, assigns joint cell types, and lifts the joint cell embedding into a single sample embedding (uns['X_DR_sample']) shared with the single-modality pipelines.

Run the wrapper, not the low-level functions

The supported entry point is the config-driven wrapper (sampledisco -m complex --config <yaml> with run_multiomics_pipeline=True), which orchestrates preparation → joint cell typing → compute_sample_embedding → downstream analysis for you. The functions below are documented for reference and for advanced step-by-step use.

Function	Purpose
multiomics_preparation	Full scGLUE pipeline — preprocessing, adversarial training, per-modality QC, embedding-union merge, optional visualization.
cell_types_multiomics	Leiden on RNA cells in the joint embedding, then Jaccard-weighted SNN label transfer to ATAC.
compute_sample_embedding	Single-key sample embedding for multi-omics (uns['X_DR_sample']); pass modality_col='modality'.
replace_optimal_dimension_reduction	Legacy optimal-resolution merge helper.
run_autotune	Parameter selection — optimizes the RMD-weight α (composition vs displacement); pass modality_col='modality'. Enabled via multiomics_autotune_enable in the wrapper.

Typical order

multiomics_preparation                  # scGLUE integration + per-modality QC + merge
   └─> cell_types_multiomics            # joint cell typing
         └─> compute_sample_embedding   # writes uns['X_DR_sample'], modality_col='modality'
               └─> (optional) multiomics_autotune_enable   # alpha / block-weight tuning
                     └─> downstream analyses

For a runnable walkthrough, see the Multi-omics tutorial.