CLIP-seq QC - Usage Guide

Overview

Comprehensive QC for CLIP libraries (eCLIP, iCLIP, iCLIP2, PAR-CLIP). Five gates: preprocessing retention -> alignment rate -> library complexity -> FRiP/IP enrichment -> replicate reproducibility (IDR). Each gate must pass for the library to be usable. ENCODE eCLIP thresholds: >= 1M unique fragments per rep, FRiP >= 0.005 (narrow RBPs), IDR rescue + self-consistency both < 2. SMInput is the canonical control - NOT IgG, NOT empty beads, NOT RNA-seq.

Prerequisites

conda install -c bioconda preseq picard samtools bedtools deeptools idr multiqc rseqc fastp
pip install pysam

Quick Start

Tell your AI agent: - "Run library complexity with preseq lc_extrap; flag if predicted unique < 1M at 100M reads" - "Compute FRiP against stringent peaks; ENCODE narrow-binding minimum 0.005" - "Run IDR true-replicate and pseudo-replicate; report rescue and self-consistency ratios" - "Read distribution metagene - HuR should be >50% 3' UTR" - "Diagnose why my CLIP library has 95% PCR duplication" - "Why is my FRiP so low - failed IP or wrong RBP class threshold?" - "Antibody sanity check: do top peaks match expected biology?"

Example Prompts

Library Complexity

"preseq lc_extrap on pre-dedup BAM; what unique-fragment count is predicted at 100M reads?"

"preseq curve flattens early - over-amplified library; nothing to do but re-prep"

FRiP

"FRiP against stringent peak set; expected 0.01-0.10 for splicing factor"

"FRiP < 0.005 - check antibody; not an atypical-binding RBP"

IDR

"IDR with --rank 5 (log2 FC) on signal-sorted replicate beds; threshold 0.05 true rep, 0.10 pseudoreplicate"

"Rescue ratio > 2 - one replicate failed; identify and re-sequence"

Diagnostics

"Top peaks are dominated by GAPDH and ACTB - failed IP or expression bias?"

"60% PCR duplication - this is normal CLIP, not low complexity"

"Read distribution shows 50% intergenic - rRNA contamination or wrong TxDb?"

Antibody Sanity

"Verify antibody on KD lysate WB before claiming the CLIP worked"

"Top motif AU-rich and GO terms generic - antibody non-specific"

What the Agent Will Do

Inspect preprocessing logs (cutadapt retention >= 70%)
STAR Log.final.out (unique alignment >= 60% eCLIP, 70% iCLIP)
preseq lc_extrap on pre-dedup BAM (>= 1M unique predicted)
FRiP calculation against stringent peaks (>= 0.005 narrow-binding)
log2(IP/SMInput) in-peak enrichment check
IDR true-replicate and pseudo-replicate; ENCODE rescue/self-consistency rules
RSeQC read_distribution + geneBody_coverage for metagene QC
Antibody sanity: top peaks match expected RBP biology
MultiQC for consolidated report

Tips

Five gates in order. Don't skip earlier ones.
40-70% duplication is normal CLIP. UMI dedup recovers unique molecules.
preseq on pre-dedup BAM. Post-dedup gives nonsense.
FRiP is RBP-class-specific. Atypical-binding RBPs exempt.
IDR ranks by log2 FC or p-value, not raw score. CLIPper score is tied.
SMInput is the canonical control. Not IgG, not RNA-seq.
Antibody is the single most common failure point. 30-50% commercial "IP-grade" antibodies fail.
Validate antibody on KD lysate WB. Knockdown -> WB signal decrease is the test.
MultiQC unifies the report. Standard practice for cross-lab work.
Failed at gate 3 (complexity) = no rescue. Re-prep.

clip-seq/clip-preprocessing - Gate 1
clip-seq/clip-alignment - Gate 2
clip-seq/clip-peak-calling - Gates 4-5
clip-seq/differential-clip - Differential QC
read-qc/quality-reports - General QC
read-qc/contamination-screening - Contamination
chip-seq/chipseq-qc - ChIP-seq analogue