← Về thư mục

Motif Analysis - Usage Guide

Overview

Discovers de novo motifs and tests known motif enrichment in ChIP-seq, ATAC-seq, or other peak sequences using HOMER, MEME-ChIP (STREME, CentriMo, TOMTOM, FIMO), monaLisa (regression-based binned analysis), and AME (differential enrichment). Embeds background-selection theory (GC-matched, dinucleotide-shuffled, Markov order-2, peak-flanks), summit-centering requirements, and recognizes when to use deep-learning attribution methods (TF-MoDISco on BPNet/chromBPNet) for soft motif syntax.

Prerequisites

# HOMER
conda install -c bioconda homer
perl $CONDA_PREFIX/share/homer/configureHomer.pl -install hg38 mm10

# MEME suite (5.4+ has STREME, not DREME)
conda install -c bioconda meme

# Motif databases
wget https://jaspar.genereg.net/download/data/2024/CORE/JASPAR2024_CORE_vertebrates_non-redundant_pfms_meme.txt
wget https://hocomoco12.autosome.org/final_bundle/hocomoco12/H12CORE/formatted_motifs/H12CORE_meme_format.meme

# Optional: monaLisa for regression-based binned analysis
BiocManager::install(c('monaLisa', 'JASPAR2024', 'BSgenome.Hsapiens.UCSC.hg38'))

Quick Start

Tell the agent what to do: - "Run HOMER findMotifsGenome.pl on summit-centered narrowPeak with repeat masking and 8 cores" - "Run MEME-ChIP with JASPAR 2024 motifs on peaks resized to ±100 bp around summit" - "Find motifs central to my CTCF peaks (test direct vs tethered binding via CentriMo)" - "Compare motif enrichment between gained-in-treatment and gained-in-control peaks (AME differential)" - "Use monaLisa to find motifs discriminating top-quintile log2FC peaks from bottom-quintile" - "Scan my motif of interest genome-wide with FIMO at p < 1e-5" - "My top motif is the AluY consensus; diagnose and fix the background"

Example Prompts

Standard discovery (TF or sharp histone)

"Run HOMER de novo + known motif discovery on FOXA1 ChIP-seq peaks. Recenter to ±100 bp around summit, mask repeats, use 8 cores. Report top 10 known motifs and top 5 de novo."

MEME-ChIP comprehensive

"Run MEME-ChIP on H3K27ac peaks with JASPAR 2024 vertebrates database. Include STREME for short motifs, MEME for longer motifs, CentriMo for central enrichment, TOMTOM for known-motif matching."

Differential motifs

"I have peaks gained in EZH2-inhibitor treatment vs DMSO control. Run AME to find motifs enriched in gained-only peaks."

Binned monaLisa

"Split my peaks into quintiles by log2FC from differential analysis. Run monaLisa to find which TF motifs discriminate top quintile from bottom, controlling for GC content."

Central enrichment test

"Run CentriMo on my CTCF peaks to verify direct sequence-specific binding (motif should be centrally enriched within ±50 bp of summit) vs tethered/indirect binding (no central enrichment)."

FIMO scanning

"Scan the human genome (hg38) for FOXA1 motif occurrences at p < 1e-5 with HOCOMOCO v12 PWM."

Diagnosing artifacts

"My top motif is GC-rich, doesn't match any TF in TOMTOM, and looks like an Alu fragment. Check for repeat contamination and re-run with -mask plus GC-matched background."

Deep-learning motifs

"Extract motifs from a trained chromBPNet model using TF-MoDISco from the attribution scores."

What the Agent Will Do

1. Validate peak input: confirm peaks pass FRiP and aren't dominated by hyper-ChIPable artifacts (see chipseq-qc)
2. Recenter peaks on summit: narrowPeak column 10 gives summit offset; resize to ±100-250 bp (sharp marks) or use full width (variable marks)
3. Extract peak sequences: bedtools getfasta with the matching genome FASTA
4. Choose background appropriately: GC-matched genomic (HOMER default with -mask), Markov order-2 (STREME / MEME-ChIP default), peak-flanks (when peak GC differs from genome), or differential set (AME)
5. Run discovery + enrichment:
6. HOMER: de novo + known in one command
7. MEME-ChIP: STREME + MEME + CentriMo + TOMTOM + FIMO
8. monaLisa: binned regression with GC control
9. AME: differential between two sequence sets
10. TOMTOM motif comparison: verify de novo motifs against JASPAR / HOCOMOCO / known TF databases
11. Verify against expected biology: known TF should rank highly for validated ChIP; failure to recover known motif is a QC failure indicator
12. Output: HTML report (HOMER / MEME-ChIP) + ranked motif table + sequence logos
13. Document: exact peak input (count, width, summit-centered or not), background choice, motif DB version, significance threshold

Decision Tree: Which Tool

Goal	Tool
Quick discovery + known + report	HOMER findMotifsGenome.pl
Comprehensive (de novo + central + DB + scan)	MEME-ChIP
Long or gapped motifs	MEME (classical, 5.4+ still works)
Short motifs (3-15 bp) at scale	STREME
Central enrichment of known motifs	CentriMo
Differential between two peak sets	AME
Binned regression (correlate with continuous variable)	monaLisa
Genome-wide motif scanning	FIMO (tighten p ≤ 1e-5)
Soft motif syntax / cooperativity	TF-MoDISco on BPNet/chromBPNet attribution (see chip-deep-learning)
Position-specific motifs (TF dimer arrangements)	STREME + CentriMo + ME-Aria

Tips

• Always summit-center sequences. Motif enrichment improves dramatically when sequences are 200 bp around summit vs full peak width. CentriMo requires summit centering.
• Mask repeats unless explicitly studying TE-derived enhancers. Repeat-derived motifs (AluY, LINE, LTR) are common false positives.
• Use Markov order-2 background for de novo (STREME/MEME default). Order-0 (mononucleotide) misses dinucleotide biases like CpG.
• HOMER's auto background can include peaks. For large peak sets (>5% of genome), supply explicit -bg.
• FIMO genome-wide at default p ≤ 1e-4 = millions of false positives. Tighten to 1e-5 for whole-genome scans; restrict to peaks for fine-grained p.
• Differential motif analysis is more interpretable than absolute enrichment. Compare gained-in-A vs gained-in-B (AME) rather than each set vs genome.
• monaLisa for continuous variables. With a per-peak measurement like log2FC, accessibility, or methylation, binned regression is more powerful than absolute enrichment.
• CentriMo central enrichment is the gold standard for direct binding. Tethered (indirect) binding shows motif enrichment but not central enrichment.
• Failure to recover the known TF motif for a validated ChIP is a QC failure. Re-check antibody validation, peak quality, and background choice before claiming "novel motif."
• For broad histone marks (H3K27me3, H3K9me3), motif analysis is generally not informative: Polycomb and heterochromatin do not depend on sequence-specific binding directly.

Troubleshooting

No significant motifs

Causes in order of frequency: 1. Peak quality / hyper-ChIPable contamination -> see chipseq-qc, filter blacklist + custom hyper-ChIPable 2. Too few peaks (< 500) -> motif discovery needs ≥ 500-1000 high-confidence peaks 3. Wrong sequence window -> recenter to summit ±100-250 bp 4. Wrong background -> for unusual GC composition, supply matched background explicitly 5. Broad histone mark -> motif analysis may not apply

Top motif is GC-rich or matches no TF database

1. Repeat contamination -> use -mask (HOMER) or pre-mask sequences
2. CpG island bias -> matched background; consider order-2 Markov
3. Hyper-ChIPable artifacts -> filter peaks; see chipseq-qc
4. Compositional control failed -> switch tool (STREME order-2 typically better)

Known TF motif not recovered

1. ChIP quality issue -> check FRiP, NSC, antibody validation
2. Tethered (indirect) binding -> CentriMo may show no central enrichment despite overall enrichment
3. Cofactor binding (motif belongs to partner) -> check known interaction partners

Too many motif hits (FIMO)

Default --thresh 1e-4 generates millions of false positives genome-wide. Use --thresh 1e-5 for whole-genome OR restrict to peaks fimo motif.meme peaks.fa.

MEME-ChIP "sequences too short"

MEME requires sequences ≥ motif min width. Resize peaks to ≥ 200 bp.

MEME-ChIP / STREME memory failure

Long sequences (> 1 kb each) at large counts exceed RAM. Resize to ±100-250 bp around summit; downsample peak count if needed.

Related Skills

• chip-seq/peak-calling - Generate input peaks
• chip-seq/chipseq-qc - Filter hyper-ChIPable peaks before motif analysis
• chip-seq/chip-deep-learning - TF-MoDISco on BPNet/chromBPNet for sequence syntax
• chip-seq/peak-annotation - Separate promoter vs enhancer peaks for stratified motif analysis
• atac-seq/motif-deviation - chromVAR per-cell motif activity (ATAC-specific)
• atac-seq/footprinting - TOBIAS footprint analysis
• sequence-manipulation/motif-search - General motif scanning
• genome-intervals/proximity-operations - Extract peak FASTA via bedtools getfasta