Splicing Pipeline - Usage Guide

Overview

Complete bulk short-read alternative splicing analysis workflow from raw RNA-seq FASTQ files to differential splicing results and visualizations. Includes QC checkpoints and best practices. For variant-driven splice prediction use splice-variant-prediction; for rare-disease single-patient outlier detection use outlier-splicing-detection; for PacBio/ONT full-isoform analysis use long-read-splicing.

Prerequisites

# Core tools
conda install -c bioconda star rmats-turbo ggsashimi fastp rseqc

# Python dependencies
pip install pandas numpy matplotlib

# R packages (optional, for isoform switching)
BiocManager::install('IsoformSwitchAnalyzeR')

Quick Start

Tell your AI agent what you want to do: - "Run a complete splicing analysis from my FASTQ files" - "Set up a splicing pipeline for my RNA-seq experiment" - "Analyze differential splicing between my conditions" - "Generate sashimi plots for my significant splicing events"

Example Prompts

Full Pipeline

"I have RNA-seq FASTQs from 3 control and 3 treatment samples. Run a complete splicing analysis."

"Set up STAR 2-pass alignment and rMATS differential splicing for my samples."

Individual Steps

"Align my samples with STAR 2-pass mode for optimal junction detection."

"Run junction saturation QC to check if I have enough sequencing depth."

"Visualize the top 20 differential splicing events with sashimi plots."

Pipeline-level decisions

"Do I need one shared junction database for the cohort or per-sample 2-pass?"

"Should I test event-level (rMATS) or isoform-level (DTU), and how do I reconcile them?"

"How do I import for DTU - why is it different from my gene-level RNA-seq import?"

"My transcript-level FDR looks inflated - what's the stageR two-stage test for?"

What the Agent Will Do

1. Perform read QC and adapter trimming (fastp)
2. Align with STAR 2-pass mode for junction detection
3. Run junction saturation QC checkpoint
4. Perform differential splicing analysis (rMATS-turbo)
5. Filter significant events (FDR < 0.05, |deltaPSI| > 0.1)
6. Optionally analyze isoform switches
7. Generate sashimi plots for visualization

Tips

• STAR 2-pass must be cohort-consistent: pass 2 uses ONE combined junction DB from all samples, or PSI is not comparable across samples
• Check junction saturation curves before trusting results
• Use --outSJfilterOverhangMin 8 8 8 8 to relax STAR's default (30 12 12 12) and retain shorter-overhang novel junctions; noise is removed later by the >=3 unique-read filter on the shared junction DB
• rMATS-turbo combines quantification and differential testing; --readLength must match the trimmed reads
• The DTU branch imports the OPPOSITE way from gene-level RNA-seq: txOut=TRUE + countsFromAbundance="dtuScaledTPM"
• Transcript-level FDR is only honest through the stageR gene->transcript two-stage test
• Standard thresholds: |deltaPSI| > 0.1, FDR < 0.05, and >= 10 supporting junction reads per event

Related Skills

• alternative-splicing/splicing-quantification - PSI computation, event taxonomy
• alternative-splicing/differential-splicing - Tool selection, MAJIQ V3, Shiba, leafcutter
• alternative-splicing/isoform-switching - DTU and NMD/domain consequences
• alternative-splicing/sashimi-plots - Visualization tools
• alternative-splicing/splicing-qc - QC prerequisites
• alternative-splicing/single-cell-splicing - Chemistry-first single-cell decision
• alternative-splicing/splice-variant-prediction - Variant impact (SpliceAI/Pangolin)
• alternative-splicing/outlier-splicing-detection - Rare-disease single-patient (FRASER 2.0)
• alternative-splicing/long-read-splicing - Full-isoform PacBio/ONT analysis
• read-alignment/star-alignment - STAR 2-pass cohort-style configuration