muon — Multi-Modal Single-Cell Analysis

Overview

muon is a Python framework for multi-modal single-cell data analysis that extends the AnnData ecosystem. Its core data structure, MuData, holds multiple AnnData objects (one per modality: RNA, ATAC, protein, etc.) with shared observation and variable axes, enabling coordinated operations across all modalities. muon provides modality-specific preprocessing routines (TF-IDF and LSI for ATAC, CLR normalization for surface proteins), Weighted Nearest Neighbor (WNN) graph construction for joint dimensionality reduction, and cross-modal analysis tools. It integrates directly with scanpy, scvi-tools, and MOFA+ for a complete multi-omics single-cell workflow.

When to Use

• Analyzing 10x Genomics Multiome data (simultaneous RNA + ATAC from the same nuclei)
• Processing CITE-seq experiments (RNA + surface protein from the same cells)
• Building joint UMAP embeddings that integrate signals from two or more modalities via WNN
• Preprocessing ATAC-seq modalities (TF-IDF normalization, LSI dimensionality reduction)
• Normalizing surface protein data with centered log-ratio (CLR) normalization
• Performing cross-modal feature linkage (associating ATAC peaks with nearby gene expression)
• Applying MOFA+ factor analysis across multiple omics layers within a unified container
• Use scanpy-scrna-seq instead when analyzing a single RNA-seq modality without any co-measured omics
• Use scvi-tools (MultiVI / totalVI) when you need probabilistic deep generative batch correction across modalities

Prerequisites