← Về thư mục
📄 / / profiles / r-and-d / raw_evidence / OpenAI4S / skills / bioskills / bio-chip-seq-super-enhancers / references / usage-guide.md

Super-Enhancers - Usage Guide

Overview

Identify super-enhancers (SE) from H3K27ac, MED1, or BRD4 ChIP-seq using ROSE / ROSE2 (canonical), LILY (input-subtraction variant), HOMER -style super, or custom hockey-stick analysis. Handles peak stitching, TSS exclusion, hockey-stick inflection, marker choice (H3K27ac vs MED1/BRD4 for BET-inhibitor predictions), spike-in normalization for cross-condition comparison, ENCODE dELS cross-referencing, and core regulatory circuitry (CRC) reconstruction. SE are operationally defined regulatory regions driving cell identity and cancer biology; see Pott & Lieb 2015 for the continuum critique.

Prerequisites

# ROSE (Python 3). For hg38 data use the stjude/ROSE fork (its genomeDict includes HG38);
# run ROSE_main.py from the cloned repo dir so it finds its annotation/ files.
git clone https://github.com/stjude/ROSE.git
# rose2 is pip-installable but its released genomeDict has no HG38 (hg18/hg19/mm*/rn* only):
# pip install rose2

# HOMER alternative
conda install -c bioconda homer

# General utilities
conda install -c bioconda samtools bedtools

# LILY (input-subtraction variant; optional)
# https://github.com/BoevaLab/LILY

Quick Start

Tell the agent what to do: - "Run ROSE2 on H3K27ac peaks with default stitching (12.5 kb) and TSS exclusion (2.5 kb)" - "Compare SE between DMSO and BET-inhibitor treatment using spike-in normalization" - "Use BRD4 ChIP for SE calling because I want to predict BET-inhibitor response" - "Build core regulatory circuitry from SE-encoded TFs" - "Cross-reference my SE against the ENCODE dELS atlas" - "My H3K27ac is low-quality; try LILY with input subtraction" - "Hockey-stick inflection looks wrong; verify ROSE2 output and adjust if needed"

Example Prompts

Standard SE calling

"Run ROSE2 on H3K27ac peaks from K562 cells. Default 12.5 kb stitching, 2.5 kb TSS exclusion, with matched input control."

BET-inhibitor responsiveness

"I want to predict which SE will lose signal under JQ1 treatment. Use BRD4 ChIP for SE calling, not H3K27ac."

Cross-condition (treatment vs control)

"Compare SE between EZH2-inhibitor-treated and DMSO H3K27ac. Apply Drosophila spike-in normalization first because EZH2i changes H3K27ac globally (indirect; via H3K27me3 loss)."

CRC reconstruction

"Run the Saint-André 2016 core regulatory circuitry algorithm: identify TFs encoded by SE-associated genes whose motifs appear in their own SE and in other SE-encoded TFs."

dELS cross-reference

"Cross-reference my SE constituents against ENCODE dELS BED. What fraction overlap? Identify SE constituents that are NOT in dELS as potentially novel."

Differential SE

"Don't call SE per condition then intersect. Instead, build the union SE set, quantify signal per condition at each, and run DiffBind with appropriate normalization."

Low-quality data

"H3K27ac FRiP is 4% (borderline). Use LILY with input subtraction instead of ROSE2."

What the Agent Will Do

  1. Choose marker: H3K27ac for discovery; MED1 or BRD4 for BET-inhibitor predictions; H3K27ac+MED1 intersection for gold-standard
  2. Pre-filter peaks: remove blacklist; remove hyper-ChIPable (top-1% input signal); remove promoter peaks if not using ROSE TSS exclusion
  3. Convert to ROSE GFF format if needed (column order: chr, source, feature, start, end, ., strand, ., ID=N)
  4. Run ROSE2 with -s 12500 -t 2500 defaults; provide input control for background subtraction
  5. Inspect hockey-stick plot: verify clean inflection; if degenerate (insufficient peaks), document and consider top-N% by signal as alternative
  6. For cross-condition comparison: spike-in normalize signal first, build union SE set, quantify at union with DiffBind
  7. For CRC: Extract SE-encoded TFs, scan their motifs across SE, build network graph, identify connected components
  8. Cross-reference dELS if ENCODE cCRE atlas available
  9. Assign target genes: nearest-TSS is naive; use ENCODE-rE2G or ABC for cell-type-specific (see chip-seq/peak-annotation)
  10. Output: SE BED, hockey-stick plot, ranked SE table, CRC network (if requested), differential SE if cross-condition
  11. Document: marker choice, stitching/TSS-exclusion parameters, normalization (spike-in if cross-condition), assigned-gene method

Tips

Troubleshooting

Too few SE called

  1. Hockey-stick inflection failed -> check _Plot_points.png; need ≥ 5000 enhancer peaks
  2. Stitching distance too small for biology -> try -s 15000 or -s 25000 for tissue samples
  3. TSS exclusion too aggressive -> reduce -t 500 if including promoter-proximal SE
  4. Input control too strong -> consider running without input or with LILY's variant

Too many SE called

  1. Hyper-ChIPable artifacts inflated input -> filter blacklist + custom top-1% input signal
  2. Insufficient differentiation in signal distribution -> hockey-stick fails; check _Plot_points.png
  3. Threshold not biologically meaningful -> use top 5-10% by signal as alternative

SE counts differ between conditions

This is expected when: - Real biology (gain/loss of regulatory programs) - Global signal shift (HDACi, BETi, EZH2i) confounding threshold - Different sequencing depth between conditions

Fix the confounding before interpreting: spike-in normalize and quantify on union SE set.

CRC network has too many components / too few

  1. Too many TFs (default lists may include non-master TFs) -> restrict to TFs with motifs in own SE
  2. Too few cross-bindings -> check motif quality (use HOCOMOCO v12 not JASPAR for robust calls)
  3. Cell type with no master regulators (rare) -> CRC may not apply

Differential SE all "novel" gain/loss

Without spike-in normalization, this is often artifact. With spike-in, expect ~30-50% conserved SE between conditions and a smaller "differential" set.