CLIP-seq Preprocessing - Usage Guide

Overview

Turn raw CLIP-seq FASTQ (eCLIP, iCLIP, iCLIP2, iCLIP3, irCLIP, PAR-CLIP, FLASH) into UMI-deduplicated, alignment-ready FASTQ. The defining constraint of CLIP preprocessing is preservation of the 5' read end where the reverse transcriptase truncated at the protein-RNA adduct: that base is one nucleotide downstream of the crosslink and is the foundation of all single-nucleotide-resolution analyses. Quality trimming and adapter trimming must operate on the 3' end only.

Prerequisites

conda install -c bioconda umi_tools cutadapt fastp samtools bowtie2 preseq picard
pip install pysam

Quick Start

Tell your AI agent what you want to do: - "Preprocess my eCLIP data with 10 nt UMIs" - "Demultiplex multiplexed iCLIP2 with the NNNXXXXNN pattern" - "Run two-pass adapter trimming for read-through events" - "Dedupe by UMI using ENCODE convention" - "Estimate library complexity with preseq" - "Preprocess PAR-CLIP without losing T-to-C mutations"

Example Prompts

Protocol-Specific UMI Extraction

"Extract 10-nt UMIs from R1 of my paired-end eCLIP library"

"My iCLIP2 has the NNNXXXXNN barcode pattern - demultiplex by the 4-base library barcode first, then extract the surrounding 5 nt as UMI"

"PAR-CLIP from the Tuschl lab - 4 nt random barcode; verify pattern by inspecting first 12 bases of reads"

Adapter Trimming

"Run cutadapt with -q 6 -m 18 and only the 3' adapter (-a/-A), never 5' (-g/-G on R1)"

"Two-pass eCLIP trimming: pass 1 is standard 3' adapter, pass 2 strips the read-through 5' adapter from R2 only"

"Raise STAR mismatch ceiling to 0.07 for PAR-CLIP so T-to-C reads are not filtered as errors"

Deduplication and Complexity

"After alignment, dedupe with umi_tools dedup --method=unique per ENCODE convention"

"Estimate library complexity with preseq lc_extrap; flag if predicted unique < 1M at 100M reads"

"Fall back to picard MarkDuplicates if my older library has no UMIs"

Reconciliation

"Compare retained read counts between fastp and cutadapt and pick the one matching ENCODE eCLIP"

"My library has 95% PCR duplication - is this normal CLIP, or did the IP fail?"

"Pre-map to rRNA before genome alignment to avoid STAR multi-mapper tangles"

What the Agent Will Do

Inspect the library prep documentation OR the first 12 bases of reads to identify protocol and UMI pattern
Run umi_tools extract with the correct --bc-pattern (10 nt eCLIP R1; NNNXXXXNN iCLIP/iCLIP2; demultiplexed UMI for iCLIP3; 4 nt PAR-CLIP)
For iCLIP2 / multiplexed libraries: demultiplex by the inline library barcode BEFORE UMI extraction
Run cutadapt with the matching 3' adapter (and 5' adapter on R2 only for two-pass eCLIP) at -q 6 -m 18
(Optional) Pre-map to rRNA with bowtie2 to remove the abundant ribosomal background
Align with STAR (ENCODE parameters - see clip-alignment skill)
UMI-dedupe with umi_tools dedup --method=unique
Compute library complexity with preseq and report ENCODE compliance

Tips

The 5' end is sacred. R2 5' end in paired-end eCLIP is the crosslink site -1. Never quality-trim or adapter-trim there.
eCLIP uses 10 nt UMIs on R1. Older protocols used 5 or 7 nt; verify before assuming 10.
iCLIP/iCLIP2 UMI is NNNXXXXNN (3+4+2). The 4 middle bases are sample barcodes; demultiplex first, then dedupe the 5 surviving Ns.
PAR-CLIP needs higher mismatch tolerance. T->C mutations are SIGNAL, not error. Raise --outFilterMismatchNoverLmax from 0.04 to 0.07.
CLIP duplication rates are 40-70% normally. The IP enriches a small molecule pool. Low duplication means failed IP, not a good library.
umi_tools dedup --method=unique is the ENCODE default. Directional clustering is more conservative but adds runtime.
18 nt minimum is mandatory. Reads < 18 nt multi-map at > 50% rates.
Library complexity is the true QC. preseq predicts plateau; ENCODE eCLIP needs >= 1M unique at sequenced depth.
Pre-map rRNA is performance, not biology. Skipping it costs ~30 minutes of STAR time but does not change downstream peak calling if done correctly.

clip-seq/clip-alignment - Downstream alignment with ENCODE parameters
clip-seq/clip-qc - QC of the preprocessed library
clip-seq/crosslink-site-detection - Why 5' preservation matters
clip-seq/stamp-antibody-free - STAMP uses RNA-seq preprocessing, not CLIP preprocessing
read-qc/umi-processing - General UMI handling
read-qc/adapter-trimming - General adapter trimming patterns