bio-chipseq-motif-analysis

What this skill does

## Version Compatibility

Reference examples tested with: HOMER 4.11+, MEME suite 5.5+ (STREME replaces DREME from 5.4+), monaLisa 1.10+, JASPAR 2024 CORE, HOCOMOCO v12, BioPython 1.83+, bedtools 2.31+.

DREME was removed from MEME suite 5.4+; use STREME instead. Some tutorials still reference DREME — verify the installed version via `meme --version`. JASPAR 2026 (released late 2025) integrates 1259 BPNet ChIP models in a Deep Learning collection; the CORE collection remains the standard PWM source.

# Motif Analysis on ChIP-seq Peaks

**"Find enriched DNA binding motifs in my ChIP-seq peaks"** -> Discover de novo motif patterns and test for known TF motif enrichment in peak sequences, with appropriate background to control for compositional and positional biases.

- CLI (HOMER, fast): `findMotifsGenome.pl peaks.bed hg38 outdir/ -size 200 -p 8`
- CLI (MEME-ChIP, comprehensive): `meme-chip -db JASPAR.meme peaks.fa`
- R (regression-based, selective enrichment): `monaLisa::calcBinnedMotifEnrR(seqs, bins, pwms)`
- CLI (deep-learning-derived motifs): TF-MoDISco on BPNet attribution scores (see chip-deep-learning)

Motif discovery is sensitive to background choice and peak quality. Hyper-ChIPable artifacts at rRNA / housekeeping loci often produce false-positive motifs (GC-rich or A-T-rich biases of those regions). Filter peaks against blacklists and inspect peak distribution before running motif discovery.

## Tool Taxonomy

| Tool | Discovery type | Background handling | Strength | Fails when |
|------|----------------|---------------------|----------|------------|
| **HOMER findMotifsGenome.pl** | De novo + known | GC-matched genomic regions (auto) | Fast (multi-core); integrated vertebrate/insect/plant DBs; one-command full report | Background can include unmasked repeats producing motif artifacts; `-size given` slow; auto background may include peaks themselves |
| **MEME-ChIP** | De novo (STREME, MEME) + central enrichment (CentriMo) + DB comparison (TOMTOM) + scanning (FIMO) | Markov order-2 from input; shuffled (preserves dinucleotide) | Comprehensive single command; rigorous statistics; HTML report | Slower; sequences must be 100-500 bp; central enrichment requires summit-centered peaks |
| **STREME** (MEME 5.4+) | De novo (replaced DREME) | Markov order-2 | Bailey 2021 benchmark: more accurate than DREME/HOMER/MEME/Peak-motifs; handles 3-30 bp; scales to 100k+ sequences | Memory-hungry for very long sequences (>1 kb) |
| **MEME** (classical) | De novo (long, gapped) | Markov | Long motifs; gapped motifs | Slow (no parallel); replaced by STREME for short motifs |
| **DREME** | De novo (short) | Shuffled | Historical; small fast | Removed from MEME 5.4+; use STREME |
| **monaLisa** (Stadler lab) | Binned enrichment regression | Native (binned scoring) | Modern; regression-based; selectivity (TF-specific in differential peaks) | R-only; less integrated with browsers |
| **AME** (MEME suite) | Known motif differential | Matched background set required | Designed for two-set comparison (e.g., peaks vs. control regions) | Requires user-provided background set |
| **CentriMo** | Known motif central enrichment | Auto from input | Tests positional enrichment relative to peak center | Requires summit-centered peaks (200-500 bp) |
| **FIMO** | Motif scanning | Markov model | Genome-wide scanning at user-set p-value | Many false positives at p ≤ 1e-4; tighten to 1e-5 for whole-genome |
| **HOMER scanMotifGenomeWide.pl** | Motif scanning | None | Genome-wide scanning at fixed score threshold | Less calibrated than FIMO; HOMER's PWM format |
| **RSAT peak-motifs** | De novo + known | k-mer comparison | Web-server; multi-tool ensemble | Web limits; less reproducible from CLI |
| **TF-MoDISco** | DL attribution-based | Implicit in model | Motifs from BPNet/chromBPNet attribution scores; captures soft motif syntax | Requires trained DL model; see chip-deep-learning |

## Background Selection — The Biggest Source of Error

Motif enrichment p-values are conditional on the background distribution. Wrong background produces wrong motifs.

| Background | What it preserves | Use case | Limitation |
|------------|-------------------|----------|------------|
| **GC-matched genomic regions** | Mononucleotide composition; chromatin context | TF motifs; avoid GC-bias artifacts | Doesn't preserve dinucleotide (CpG, TpA) |
| **Dinucleotide-shuffled** | CpG and TpA frequencies | Short motifs; avoiding repeat-derived artifacts | Doesn't capture genomic position context |
| **Markov order-2 (trinucleotide)** | Trinucleotide context | Compositional control; STREME/MEME default | Doesn't capture chromatin context |
| **Peak-flanking sequences** (±500 bp upstream/downstream of peak) | Local genomic context | When peak GC differs from genome | May contain shared regulatory motifs if peaks cluster |
| **Repeat-masked input** | Sequence with repeats replaced by N | Avoid TE-derived motif artifacts | Loses motifs in evolved-from-repeat regulatory elements |
| **Input control peaks** | Open-chromatin / artifact regions | TF discrimination from generic chromatin | Hard to obtain; controversial |
| **Differential set** (AME) | Treatment-condition-specific peaks vs ctrl peaks | Differential motif enrichment | Requires a control peak set |

**Practical default:** STREME / MEME-ChIP with Markov order-2 (built-in default); HOMER with `-mask` flag (mask repeats); always inspect for repeat-derived motifs (e.g., Alu-derived AluY consensus, LINE motifs).

## Window Around Summit Matters

Motif enrichment improves dramatically when sequences are summit-centered:

| Window | When |
|--------|------|
| ±100 bp (200 bp total) | Sharp TFs (CTCF, p53); summit reliably reflects motif position |
| ±150-250 bp (300-500 bp) | Most TFs and sharp histones; balance of motif coverage and noise |
| Full peak width (`-size given`) | Variable-width broad marks; computationally expensive |
| Whole gene body (>1 kb) | Almost always wrong; dilutes motif signal |

**For ChIP-seq broad histone marks (H3K27me3, H3K9me3) motif analysis is generally NOT informative** — these marks reflect Polycomb / heterochromatin domains without sequence-specific binding. Motif analysis applies to TFs and to histone marks deposited by sequence-specific cofactors (H3K27ac partial, since BRD4 reads acetyl).

## HOMER Workflow

```bash
# De novo + known motif discovery, repeat-masked, GC-matched background
findMotifsGenome.pl peaks.narrowPeak hg38 homer_out/ \
    -size 200 \
    -mask \
    -p 8

# With user-supplied background (e.g., control peaks or random genomic)
findMotifsGenome.pl peaks.narrowPeak hg38 homer_out/ \
    -size 200 -mask -p 8 \
    -bg background_peaks.bed

# Known motifs only (skip de novo; faster)
findMotifsGenome.pl peaks.narrowPeak hg38 homer_known_only/ \
    -size 200 -mask -nomotif

# Differential motif analysis: peaks gained in condition A vs condition B
findMotifsGenome.pl gained_in_A.bed hg38 differential_motifs/ \
    -size 200 -mask -bg gained_in_B.bed
```

HOMER output files:
- `homerResults.html` — de novo motifs ranked by significance
- `knownResults.html` — known motif enrichment
- `homerMotifs.all.motifs` — all de novo motifs (PWM format)
- `knownResults.txt` — tab-separated known motif stats

## MEME-ChIP Workflow

```bash
# Center peaks to ±100 bp around summit (column 10 in narrowPeak)
awk 'BEGIN{OFS="\t"} {summit = $2 + $10; print $1, summit - 100, summit + 100, $4, $5, $6}' \
    peaks.narrowPeak > peaks_centered.bed
bedtools getfasta -fi hg38.fa -bed peaks_centered.bed -fo peaks_centered.fa

# Full MEME-ChIP analysis
meme-chip \
    -oc meme_chip_out/ \
    -db JASPAR2024_CORE_vertebrates_non-redundant_pfms_meme.txt \
    -meme-nmotifs 5 \
    -streme-nmotifs 10 \
    -minw 6 -maxw 20 \
    peaks_centered.fa

# MEME-ChIP runs: STREME (replaces DREME) + MEME + CentriMo + TOMTOM + FIMO
# CentriMo tests known motifs for central enrichment — strongest signal of
# direct binding vs. tethered/indir