Single-Cell RNA-seq Pipeline - Usage Guide

Overview

This workflow processes single-cell RNA-seq data from 10X Genomics Cell Ranger output to annotated cell types. It supports both Seurat (R) and Scanpy (Python) implementations.

Prerequisites

# R packages
install.packages(c('Seurat', 'ggplot2', 'dplyr'))
BiocManager::install(c('scDblFinder', 'SingleCellExperiment'))

# Python packages
pip install scanpy scrublet anndata

Quick Start

Tell your AI agent what you want to do: - "Analyze my 10X single-cell data from start to finish" - "Run the scRNA-seq pipeline on my PBMC data" - "Cluster my single-cell data and find marker genes"

Example Prompts

Starting from Cell Ranger output

"I have filtered_feature_bc_matrix from Cell Ranger, analyze it"

"Load my 10X data and perform QC filtering"

"Process my scRNA-seq with Scanpy instead of Seurat"

Analysis steps

"Remove doublets from my single-cell data"

"Find clusters at different resolutions"

"What are the marker genes for cluster 3?"

Annotation

"Help me annotate cell types based on marker genes"

"Use SingleR for automated cell type annotation"

Input Requirements

Input	Format	Description
10X data	filtered_feature_bc_matrix/	Directory with matrix.mtx, barcodes.tsv, features.tsv
10X H5	.h5	Alternatively, the HDF5 format

What the Workflow Does

Load Data - Read 10X matrix into Seurat/AnnData
QC Filtering - Remove low-quality cells and doublets
Normalization - SCTransform or log-normalization
Feature Selection - Find highly variable genes
Dimension Reduction - PCA and UMAP
Clustering - Graph-based clustering
Markers - Find cluster-specific genes
Annotation - Assign cell type labels

Seurat vs Scanpy

Feature	Seurat	Scanpy
Language	R	Python
Speed	Fast	Faster for large data
Memory	Moderate	Lower
Ecosystem	Bioconductor	Python ML stack
Best for	General analysis	Large datasets, integration

Tips

Cell numbers: Expect 1,000-20,000 cells from typical 10X run
Genes per cell: 200-5,000 is typical; very high may be doublets
Mitochondrial: >20% suggests dying cells
Resolution: Start at 0.5, adjust based on cluster quality
Annotation: Check canonical markers for your tissue type
Stage order: QC and doublet removal run per sample BEFORE integration; integrate, then cluster, then annotate, never reorder
Multi-sample designs: Batch-correct on a shared embedding (Harmony/scVI/RPCA) and watch for over-correction; batch-mixing metrics alone reward flattening real biology, so pair them with a bio-conservation metric
Condition DE: Aggregate raw counts to pseudobulk per sample and cell type, then use DESeq2/edgeR; cluster-marker p-values are descriptive labels, not condition inference
Composition vs expression: A "DE" signal between conditions can be a proportion shift, so pair condition DE with a differential-abundance test (Milo or scCODA/sccomp/propeller)

single-cell/preprocessing - QC thresholds, ambient-RNA removal, normalization choice
single-cell/doublet-detection - Per-sample doublet calling before integration
single-cell/hashing-demultiplexing - Demultiplex hashed/multiplexed pools to samples before QC
single-cell/batch-integration - Multi-sample integration and over-correction diagnosis
single-cell/clustering - Resolution sweep and cluster validation
single-cell/markers-annotation - Marker discovery, manual labeling, and pseudobulk condition DE
single-cell/cell-annotation - Automated reference-based label transfer
single-cell/differential-abundance - Test whether cell-type proportions shifted between conditions
single-cell/trajectory-inference - Pseudotime and lineage reconstruction for continuous processes
differential-expression/deseq2-basics - Pseudobulk condition DE engine for aggregated counts
pathway-analysis/go-enrichment - Functional interpretation of marker and DE gene lists