muon-multiomics-singlecell

What it does

Orchestrates multi-modal single-cell analysis with MuData containers, enabling joint processing of RNA, ATAC, and other modalities. It includes per-modality preprocessing (RNA via scanpy workflow, ATAC via TF-IDF and LSI), construction of a Weighted Nearest Neighbor (WNN) graph across modalities, and downstream steps for UMAP embedding and Leiden clustering based on the integrated graph. It also covers cross-modal analyses (peak-to-gene linking, MOFA+ factor analysis) and visualization utilities for modality-aware plots.

How it works

• Uses MuData to store multiple AnnData modalities (e.g., RNA and ATAC) with shared obs/var axes.
• RNA preprocessing: quality control, normalization, log-transform, HVG selection, PCA; updates the RNA modality within MuData.
• ATAC preprocessing: QC, TF-IDF normalization, LSI dimensionality reduction; updates the ATAC modality within MuData.
• Per-modality neighbor graphs: computes modality-specific neighbors (optional) and then builds a joint WNN graph using both RNA and ATAC embeddings.
• Joint embedding and clustering: runs UMAP based on the WNN graph and clusters with Leiden using the WNN neighbors.
• Visualization: supports modality-aware plotting functions for joint embeddings and per-modality colorings.
• Cross-modal analysis: provides hooks for peak-to-gene annotation, differential peak analysis, and MOFA+ factor analysis, storing results in MuData structures.

When to use it

• Analyzing multi-omics data where RNA and ATAC (or other modalities) are measured in the same cells (e.g., 10x Multiome, CITE-seq).
• Building a joint embedding that integrates signals across modalities via WNN.
• Preprocessing ATAC with TF-IDF and LSI, or normalizing surface protein data with CLR.
• Performing MOFA+ factor analysis across modalities and exploring cross-modal features.

What it can touch

• MuData container holding modalities (e.g., RNA, ATAC) with shared obs; updates to mdata.mod["rna"] and mdata.mod["atac"].
• Per-modality data structures and embeddings (e.g., X_pca for RNA, X_lsi for ATAC) used in neighbor computations and UMAP.
• Visualization and clustering results stored in the MuData object (e.g., obs via Leiden clusters).

Caveats

• Requires Python packages muon>=0.1.6, scanpy>=1.10, anndata>=0.10, numpy, scipy, pandas, matplotlib, leidenalg; optional mofapy2 for MOFA+.
• MOFA+ analysis requires mofapy2; otherwise MOFA steps are skipped with an informative message.
• TF-IDF and LSI for ATAC focus on coordinated cross-modal analysis; first LSI component may reflect sequencing depth and is typically excluded in downstream use.