CLIP Motif Analysis - Usage Guide

Overview

Discover RBP binding motifs from CLIP-seq peaks or single-nucleotide crosslink sites. The fundamental confound is the uracil bias of UV254 crosslinking - U is the most-crosslinked base, so naive logos centered on CL positions are U-enriched even for non-U-binding RBPs. Modern CL-registered tools (mCross, PEKA) correct for this; legacy peak-based tools (HOMER, MEME, STREME) need GC-matched backgrounds. In vitro RBNS Kd from Dominguez 2018 is the orthogonal validation standard.

Prerequisites

conda install -c bioconda homer meme bedtools
# CL-registered tools
git clone https://github.com/chaolinzhanglab/mCross
pip install peka
# RBPamp
pip install rbpamp

Quick Start

Tell your AI agent what you want to do: - "Run HOMER de novo motif on my CLIPper peaks with a GC-matched 3' UTR background" - "Register motifs to crosslink positions with mCross" - "Confirm mCross result with PEKA orthogonal method" - "Compare in vivo CLIP motif to in vitro RBNS Kd (Dominguez 2018)" - "Scan known RBP motifs from CISBP-RNA / ATtRACT with FIMO" - "STREME because MEME is too slow on 5000 peaks" - "Generate position-specific logo with kpLogo"

Example Prompts

De Novo Discovery

"HOMER findMotifs.pl on my stringent peak set with -rna -len 5,6,7,8 and GC-matched 3' UTR background"

"STREME on 10000 peak sequences - MEME is too slow"

CL-Registered Motif

"Run mCross on PureCLIP single-nt sites; show me the CL-position-offset histogram and the registered PWM"

"PEKA for positional k-mer enrichment as orthogonal confirmation of mCross"

Known Motif Validation

"FIMO scan against CISBP-RNA PWM for PTBP1 at threshold 1e-5"

"Show me the top RBNS Kd-ranked motif for HNRNPK from Dominguez 2018 and compare to my CLIP top hit"

Diagnostics

"Why is my HOMER top motif all AU? Check for background mismatch"

"Inspect motif information content per position; if mean IC < 0.5 something is wrong"

"Validate the U-bias correction in my mCross output"

Reconciliation

"HOMER says one motif, RBNS says another - reconcile in vivo vs in vitro"

"Multiple modes in mCross - is this multimodal RBP (SRSF1) or noise?"

What the Agent Will Do

Extract peak sequences with bedtools getfasta -s preserving strand
Generate GC-matched background from expressed 3' UTRs / introns / CDS (not naive shuffled)
Run HOMER -rna -len 5,6,7,8 and STREME for de novo
For mCross / PEKA: provide single-nt CL sites from PureCLIP or CTK CITS
FIMO scan against CISBP-RNA / ATtRACT for known motifs (if RBP is published)
Compare to RBNS Kd (Dominguez 2018) if available
Report QC: information content per position, replicate concordance, CL-offset histogram for mCross
Reconcile: 3 independent tools agreeing on core motif = high confidence

Tips

U bias is real. CL sites are strongly enriched at U. Cross-check central column of PWM.
HOMER background matters. Auto-shuffled is GC-blind; provide GC-matched.
mCross needs single-nt CL sites. Peak BED breaks the model.
PEKA does not need input control. Internal normalization vs low-count crosslinks.
In vitro RBNS Kd is the gold reference. Dominguez 2018 = 78 RBPs.
Skipper window motifs and CLIPper peak motifs may differ. Window dilutes; peak sharpens.
STREME > MEME for large peak sets. MEME is O(N^2).
Information content > 1.0 bits per position = real motif. < 0.5 = no signal.
FIMO threshold 1e-5 for in vivo data. Default 1e-4 too lenient.
Multimode is biology. SRSF1 GGA half-sites, PTBP1 multimer; report both.

clip-seq/clip-peak-calling - Upstream peaks
clip-seq/crosslink-site-detection - Single-nt CL sites for mCross / PEKA
clip-seq/binding-site-annotation - Region context
clip-seq/clip-deep-learning - RBPNet / DeepRiPe sequence-to-binding
chip-seq/motif-analysis - DNA-protein analogue