CLIP-seq Peak Calling - Usage Guide

Overview

Call protein-RNA binding sites from a deduplicated CLIP-seq BAM. Three statistical frameworks compete: coverage-based (CLIPper, Piranha, MACS3), single-nucleotide-crosslink-based (PureCLIP, CTK CIMS/CITS, PARalyzer), and window-based (Skipper). The Yeo lab CLIPper + SMInput log2 normalization is the ENCODE canon; Skipper 2023 is the modern high-sensitivity replacement; PureCLIP delivers single-nt CL maps for motif registration and allele-specific binding. ENCODE stringent thresholds are log2(IP/SMInput) >= 3 AND -log10 p >= 3.

Prerequisites

conda install -c bioconda clipper pureclip piranha bedtools samtools idr
# Skipper - Snakemake workflow
git clone https://github.com/algaebrown/skipper && cd skipper && conda env create -f environment.yaml
# CTK
git clone https://github.com/chaolinzhanglab/ctk
# CLAM
pip install CLAM

Quick Start

Tell your AI agent what you want to do: - "Run CLIPper on my eCLIP and normalize against SMInput per ENCODE convention" - "Apply ENCODE stringent thresholds: log2 FC >= 3, -log10 p >= 3" - "Call single-nucleotide crosslink sites with PureCLIP" - "Use Skipper for maximum sensitivity with proper input normalization" - "Rescue repeat peaks with CLAM" - "Call PAR-CLIP T-to-C sites with PARalyzer or CTK CIMS" - "Run IDR for reproducibility between my two replicates"

Example Prompts

ENCODE-comparable CLIPper workflow

"CLIPper on rep1 and rep2 of my RBFOX2 eCLIP, normalize against SMInput, filter at stringent log2 >= 3 / -log10 p >= 3, IDR threshold 0.05"

Skipper (maximum sensitivity)

"Use Skipper for FASTKD2 - mitochondrial RBP; CLIPper misses chrM peaks"

"Compare Skipper site count to CLIPper for PUM2 - expect 2-3x more sites in 3' UTRs"

Single-nt crosslink sites

"Run PureCLIP with the IP BAM and SMInput BAM; output single-nt sites for mCross motif registration"

"Get CTK CITS truncation sites for my iCLIP - this is what PureCLIP would call but using empirical FDR"

Repeat-binding RBPs

"MATR3 binds LINE-1 - run CLAM on the multi-mapper BAM from STAR with --outFilterMultimapNmax 100"

PAR-CLIP

"PARalyzer on PAR-CLIP data with T->C transitions; output high-confidence kernel-density clusters"

"CTK CIMS with -substitution T C for PAR-CLIP - empirical FDR alternative to PARalyzer"

No SMInput

"I only have IP BAM, no SMInput - call peaks with Piranha ZTNB or PureCLIP without -ibam"

Differential / replicate pooling

"Pool rep1 and rep2 CLIPper calls; IDR threshold 0.05 for true reps and 0.10 for pseudoreps"

"Skipper output is windowed - downstream Flipper for differential binding across conditions"

What the Agent Will Do

Choose the caller based on (a) CLIP variant, (b) SMInput availability, (c) RBP binding mode, (d) goal (ENCODE comparison vs sensitivity vs single-nt)
Run the caller with appropriate parameters (CLIPper --FDR 0.05 --superlocal; PureCLIP with -ibam; Skipper Snakemake; Piranha -d ZTNB -l with covariates)
For CLIPper: normalize against SMInput with the Yeo lab scripts to produce log2 FC + -log10 p
Apply ENCODE stringent or lenient thresholds
Run IDR for replicate reproducibility; ENCODE rule: max/min of Nt and Nself <= 2
Output QC: peak count, mean width, FRiP, chrM peaks, top-peak transcript distribution
Flag failure modes (housekeeping bias from Piranha, focal PureCLIP miss of broad zones, CLIPper chrM blind spot)

Tips

CLIPper alone is not "ENCODE peaks". Always normalize against SMInput.
Stringent: log2 FC >= 3 AND -log10 p >= 3. Lenient: log2 >= 1 AND -log10 >= 2. Both apply to IDR-passing peaks.
Skipper is the 2023 sensitivity king. 210-320% more sites than CLIPper for mRNA-binding RBPs; requires SMInput.
PureCLIP is focal. It emits far fewer, single-nucleotide sites than coverage callers (very focal, low recall in the Boyle 2023 benchmark). Pair with CLIPper for broad sites.
Piranha biases to high-expression. Use SMInput as covariate via -c -l or switch to Skipper.
CLIPper misses chrM. Switch to Skipper for mitochondrial RBPs (FASTKD2, LRPPRC).
PureCLIP misses broad zones. Use it for single-nt CL maps; not for broad sites.
CLAM rescues repeat peaks. Required for MATR3, ZFP36, HNRNPK at SINE/LINE, FUS at LINE-1.
CTK CIMS for HITS-CLIP/PAR-CLIP; CTK CITS for iCLIP/eCLIP. Mutation vs truncation methods.
IDR rank by signalValue or log2 FC, not raw score. CLIPper score is sparse and tied.
FRiP >= 0.005 for ENCODE narrow. Below this = failed library or atypical-binding RBP.

clip-seq/clip-preprocessing - Upstream preprocessing
clip-seq/clip-alignment - Upstream alignment
clip-seq/clip-qc - FRiP, IDR, library complexity
clip-seq/crosslink-site-detection - Single-nt CL sites
clip-seq/differential-clip - Cross-condition differential
clip-seq/binding-site-annotation - Annotate peaks
clip-seq/clip-motif-analysis - Motifs on peaks
chip-seq/peak-calling - DNA-protein comparison