Multiome Pipeline - Usage Guide

Overview

This workflow analyzes 10X Multiome data (joint scRNA-seq + scATAC-seq from the same cells) using Seurat and Signac for weighted nearest neighbor integration.

Prerequisites

install.packages(c('Seurat', 'ggplot2'))
BiocManager::install(c('Signac', 'EnsDb.Hsapiens.v86', 'BSgenome.Hsapiens.UCSC.hg38'))

Quick Start

Tell your AI agent what you want to do: - "Analyze my 10X Multiome data" - "Integrate scRNA and scATAC from the same cells" - "Run WNN clustering on my multiome experiment"

Example Prompts

Processing

"Load my multiome Cell Ranger output"

"Filter cells by RNA and ATAC QC metrics"

Analysis

"Run weighted nearest neighbors integration"

"Find gene-peak links"

"Compare RNA vs ATAC clustering"

Input Requirements

Input	Format	Description
Cell Ranger output	Directory	Multiome processed data
Fragments file	TSV.gz	ATAC fragment positions

What the Workflow Does

Load Data - Read RNA and ATAC from same cells
RNA QC - Standard scRNA-seq filtering
ATAC QC - TSS enrichment, nucleosome signal
Process RNA - SCTransform, PCA
Process ATAC - TF-IDF, LSI
WNN - Joint embedding
Linkage - Gene-peak correlations

Tips

Paired-cell anchor: Keep only barcodes passing QC in BOTH modalities; the surviving cell set is the intersection of RNA and ATAC filters
ATAC doublets: RNA-based callers miss ATAC doublets; add a fragment-based detector (AMULET) before the joint embedding
WNN vs MultiVI: WNN weights precomputed PCA + LSI per cell and needs both modalities present; MultiVI models RNA + ATAC counts jointly and handles batch and mosaic cells better
Stage order: Per-modality QC and doublet removal first, then WNN embedding, then clustering, then annotation
Annotate from RNA: Cell-type labels come from RNA markers; ATAC gene-activity scores approximate expression and inform regulatory state, not identity
Condition comparisons: Aggregate to pseudobulk per sample for DE, and test proportion shifts separately with a differential-abundance method
LSI component 1: Usually (not always) depth-correlated; confirm with DepthCor and drop whichever component correlates with depth (commonly dims=2:30)
WNN weights: Check modality contribution per cluster; ATAC sparseness can dominate noise
Gene-peak links: Signac LinkPeaks for direct correlation; for full ABC enhancer-gene see atac-seq/enhancer-gene-linking
Cell types: Annotate using RNA markers primarily; gene activity scores from ATAC are approximate
cellranger-arc vs cellranger-atac: Multiome (paired) requires cellranger-arc; barcode universes differ
Per-cell TF activity: chromVAR via Signac::RunChromVAR after AddMotifs; see atac-seq/motif-deviation
Cis-regulatory inference: Cicero on the ATAC assay or ArchR getCoAccessibility; see atac-seq/co-accessibility

single-cell/data-io - 10X, h5ad, RDS, and h5mu loading
single-cell/preprocessing - Per-modality QC and normalization
single-cell/doublet-detection - RNA-based and hashing doublet removal
single-cell/clustering - Cluster validation on the joint graph
single-cell/markers-annotation - Marker discovery and pseudobulk condition DE
single-cell/cell-annotation - Reference-based label transfer from the RNA modality
single-cell/differential-abundance - Test whether cell-type proportions shifted between conditions
single-cell/multimodal-integration - WNN, totalVI/MultiVI, and MOFA details
single-cell/scatac-analysis - ATAC-specific processing and AMULET fragment-based doublet detection
differential-expression/deseq2-basics - Pseudobulk condition DE engine for aggregated counts
atac-seq/single-cell-atac - Signac / ArchR / SnapATAC2 ecosystem decision
atac-seq/co-accessibility - Cicero / ArchR getCoAccessibility for cis-regulatory inference
atac-seq/enhancer-gene-linking - ABC, ENCODE-rE2G for enhancer-gene mapping
atac-seq/motif-deviation - chromVAR for per-cell TF motif activity
atac-seq/footprinting - scprinter for single-cell footprinting