Differential CLIP - Usage Guide

Overview

Identify regions with differential RBP binding across conditions (KD vs control, treatment vs vehicle). Three scales: window-level (DEWSeq, Flipper), peak-level (DESeq2/edgeR/limma-voom), or crosslink-level (rare). The canonical CLIP statistical design uses the interaction term ~ type + condition + type:condition testing whether the IP/SMInput ratio shifts across conditions - NOT testing condition on IP counts alone (which confounds binding with expression). DEWSeq + htseq-clip is the EMBL-Hentze framework; Flipper is the Skipper-companion modern alternative.

Prerequisites

BiocManager::install(c('DEWSeq','DESeq2','edgeR','limma'))

pip install htseq-clip
# Flipper (with Skipper)
git clone https://github.com/algaebrown/skipper

Quick Start

Tell your AI agent: - "DEWSeq + htseq-clip on eCLIP knockdown vs control with the interaction design" - "Flipper on my Skipper output for treatment vs vehicle" - "DESeq2 peak-level on consensus CLIPper peaks with ~ type + condition + type:condition" - "Aggregate adjacent significant DEWSeq windows with bedtools merge -d 100" - "Why are my DE results dominated by housekeeping genes? Forgot the interaction term" - "KD shows ~0 DE - check siRNA efficiency by WB and avoid TMM normalization" - "Allele-specific differential with BEAPR"

Example Prompts

DEWSeq Window-Level

"htseq-clip extract with 50nt windows / 20nt step; htseq-clip count on dedup BAM; DEWSeq design ~ type + condition + type:condition"

"Aggregate adjacent significant windows within 100 nt"

Flipper (Skipper-Coupled)

"Run Flipper on the Skipper output directory for treatment vs control contrast"

Peak-Level

"Generate consensus peakset across all samples with bedtools merge; featureCounts on consensus; DESeq2 with interaction term"

Design Discussions

"Interaction term type:condition is the biologically meaningful test - tells me whether IP/SMInput ratio differs across conditions"

"Don't use TMM normalization on CLIP - it forces global shifts invisible"

"Use multiple independent siRNAs and require concordance"

Diagnostics

"DE results dominated by housekeeping - missing interaction term"

"0 DE peaks in KD - check KD efficiency"

"Significant windows scattered - need bedtools merge -d 100"

What the Agent Will Do

Identify upstream peak caller (CLIPper, Skipper) and pick matching differential framework
For CLIPper upstream: DEWSeq + htseq-clip windows; or peak-level with consensus peakset
For Skipper upstream: Flipper
Design ~ type + condition + type:condition to test SMInput-aware differential binding
Apply padj < 0.05 AND |log2FC| > 1 thresholds
For window-level: aggregate adjacent significant windows
For KD experiments: validate KD efficiency; use multiple siRNAs
For rescue: re-introduce siRNA-resistant RBP cDNA

Tips

Interaction term is mandatory. ~ type + condition + type:condition tests the actual differential.
Consensus peakset across conditions. Per-condition peaks bias to "treatment-only" peaks.
SMInput-aware design. TMM normalization erases global shifts.
n >= 2 reps per condition. n=2 marginal for DESeq2 dispersion fit.
DEWSeq for CLIPper; Flipper for Skipper. Match tool to upstream caller.
Window-level captures narrow shifts. Peak-level loses sub-peak resolution.
Aggregate adjacent windows. bedtools merge -d 100 for biological regions.
Validate KD efficiency by WB. > 70% protein loss target.
Multiple siRNAs. Off-target effects vary; require concordance.
Spike-in normalization for global shifts. Drosophila S2 nuclei or similar.

clip-seq/clip-peak-calling - Upstream peaks
clip-seq/clip-qc - QC required
clip-seq/binding-site-annotation - Annotate DE regions
clip-seq/clip-motif-analysis - Motifs on DE windows
clip-seq/ago-clip-mirna-targets - miRNA target DE
differential-expression/deseq2-basics - NB GLM
differential-expression/de-results - DE interpretation
chip-seq/differential-binding - DNA-protein analogue