RNA-seq to Differential Expression - Usage Guide

Overview

This workflow takes you from raw RNA-seq FASTQ files to a list of differentially expressed genes. It combines quality control, quantification, and statistical analysis into a complete pipeline.

Prerequisites

# CLI tools
conda install -c bioconda fastp salmon star subread

# R packages
install.packages('BiocManager')
BiocManager::install(c('DESeq2', 'tximport', 'apeglm'))
install.packages(c('ggplot2', 'pheatmap', 'ggrepel'))

Quick Start

Tell your AI agent what you want to do: - "Run the RNA-seq to DE workflow on my FASTQ files" - "I have paired-end RNA-seq data, help me find differentially expressed genes" - "Process my RNA-seq from raw reads through DESeq2"

Example Prompts

Starting from FASTQ

"I have FASTQ files for 3 control and 3 treated samples, run the full RNA-seq pipeline"

"Quantify my RNA-seq with Salmon and run DESeq2"

"Use STAR alignment instead of Salmon for my RNA-seq"

Customizing the workflow

"Run the pipeline but use a custom GTF file"

"Add batch correction to my RNA-seq analysis"

"Skip the QC step, my reads are already trimmed"

From quantification to DE

"I already have Salmon quant files, just run DESeq2"

"Import my featureCounts output into DESeq2"

Pipeline-level decisions

"Which reference release should I pin, and does my Salmon index have to match the tx2gene?"

"Should I use Salmon or STAR+featureCounts for this experiment?"

"How do I set strandedness for featureCounts - my counts look collapsed"

"My samples were prepped in two batches - how do I handle it without inflating my results?"

"I want to run GSEA next - what ranking statistic should the DE table carry?"

Input Requirements

Input	Format	Description
FASTQ files	.fastq.gz	Paired-end reads (R1 and R2 per sample)
Sample metadata	CSV/TSV	Sample names and experimental conditions
Reference	FASTA + GTF	Transcriptome for Salmon, genome + GTF for STAR
tx2gene	CSV	Transcript-to-gene mapping (for Salmon)

What the Workflow Does

Commit the reference once - Pin one Ensembl/GENCODE release for BOTH the Salmon index and the tx2gene/GTF, and fix the gene-ID namespace (ENSG backbone)
Quality Control - Trim adapters and low-quality bases with fastp
Quantification - Salmon (decoy-aware) or STAR+featureCounts to the committed reference, with strandedness verified not assumed
Import - Collapse transcript->gene through tximport (carries the length offset), never by summing NumReads
Differential Expression - Run DESeq2/edgeR/limma-voom on RAW counts with batch in the design
Results - Shrink LFC for ranking (pull the Wald stat from the unshrunk fit for GSEA), visualize on VST, export the annotated table

Choosing Between Paths

Criterion	Salmon Path	STAR Path
Speed	Faster (no alignment)	Slower
Storage	Lower (no BAM files)	Higher (BAM files)
Use case	Standard DE analysis	Need BAMs for other analyses
Accuracy	Excellent for DE	Excellent for DE
Novel junctions	No	Yes (with 2-pass)

Tips

Replicates: Minimum 3 biological replicates per condition (more is better)
Sequencing depth: 20-30M reads per sample is typical for DE analysis
Decoy index: Build the Salmon index with genome decoys; a transcriptome-only index misassigns intron and pseudogene reads
Library type: Salmon auto-detects with -l A; verify the call by reading lib_format_counts.json, and set featureCounts -s to match (dUTP/TruSeq is reverse, -s 2)
Batch effects: If samples were processed in batches, include batch in the design formula, unless batch is confounded with condition (then the design is unfixable)
Outliers: Check PCA; let DESeq2 handle single-gene outliers via Cook's distance, and remove a whole-sample outlier only with a documented technical cause
Transcript-level: For differential transcript expression or usage, generate Salmon Gibbs samples and use the transcript-aware tools in alternative-splicing/isoform-switching