← Về thư mục

Spike-In Normalization - Usage Guide

Overview

Normalize ChIP-seq data using exogenous spike-in chromatin: Drosophila chromatin (ChIP-Rx; Orlando 2014 / Egan 2016) for human/mouse ChIP, or E. coli carryover from bacterial pA-MNase/pA-Tn5 for CUT&RUN/CUT&Tag. Required when global signal shifts are expected (HDACi, BETi, EZH2i, target knockdown, dosage). Covers RRPM vs Rx-Input scaling formulas, integration with DiffBind/DESeq2/edgeR/csaw, ChIPseqSpikeInFree for post-hoc detection without spike-in, and the Patel et al 2024 review's failure-mode framework. Emphasizes the internal-control sanity check (blacklist regions should NOT shift post-scaling).

Prerequisites

conda install -c bioconda samtools bowtie2 bedtools deeptools

BiocManager::install(c('DiffBind', 'DESeq2', 'edgeR', 'csaw', 'ChIPseqSpikeInFree'))

# SpikeFlow (Snakemake wrapper, optional)
# https://github.com/DavideBrex/SpikeFlow

Quick Start

Tell the agent what to do: - "I have H3K27ac ChIP-seq from DMSO vs JQ1 (BET inhibitor). Compute Drosophila spike-in scaling factors and apply via DiffBind." - "Align reads to combined hg38 + dm6 genome and count Drosophila reads per sample" - "Apply spike-in scaling at the read level via DESeq2 sizeFactors (inverse convention)" - "Generate spike-in-scaled bigWigs with --scaleFactor alone (not combined with --normalizeUsing)" - "Verify internal-control sanity: blacklist regions should show no signal change post-scaling" - "I didn't add spike-in. Run ChIPseqSpikeInFree for post-hoc detection of global shifts (diagnostic only)." - "Compute Rx-Input scaling factor incorporating spike-in fractions in BOTH ChIP and input"

Example Prompts

Standard ChIP-Rx workflow

"I have 4 samples (2 DMSO, 2 JQ1) for H3K27ac. Build a combined hg38+dm6 bowtie2 index, align reads, filter with -F 1804 -q 30, count Drosophila reads per sample, compute scale factors as min/each, and apply to DESeq2 as sizeFactors(dds) <- 1/scale_factors."

DiffBind integration

"Use DiffBind 3.20+ with dba.normalize(obj, spikein = TRUE). Sample sheet has Spikein column with Drosophila read counts."

Spike-in scaled tracks

"Generate spike-in-scaled bigWigs with bamCoverage --scaleFactor X. Do NOT also pass --normalizeUsing."

Internal-control validation

"After applying spike-in scaling, verify ENCODE blacklist regions show no signal change between conditions. Also check signal at U6 snRNA promoter (constitutively bound by Pol III; should be stable across H3K27ac conditions)."

Post-hoc detection

"I didn't run spike-in but suspect a global shift in my BET-inhibitor experiment. Run ChIPseqSpikeInFree to estimate scaling factors from signal-distribution shape."

Rx-Input variant

"Apply Rx-Input scaling (Fursova 2019) which scales by both ChIP and input spike-in fractions to correct for IP efficiency variation."

Failure-mode debugging

"My results have wrong signs after scaling. Check inverse convention; check whether scaling was applied to peak counts instead of read counts; verify internal-control regions."

What the Agent Will Do

1. Determine spike-in protocol: Drosophila ChIP-Rx (deliberate) or E. coli carryover (CUT&RUN/CUT&Tag)
2. Build combined alignment index: host + spike genome (avoid chromosome naming collisions; prefix Drosophila chromosomes if needed)
3. Align reads: bowtie2 or bwa-mem to combined genome
4. Filter and deduplicate: ENCODE filter -F 1804 -q 30 applied to both target and spike alignments
5. Count spike reads: at high mapq, post-dedup; verify counts are in linear range (0.5-5% of total reads)
6. Compute scaling factors: RRPM: min(spike) / per_sample_spike; Rx-Input: ratio of ChIP/input spike fractions
7. Apply at correct layer:
8. bigWig tracks: bamCoverage --scaleFactor X (NOT with --normalizeUsing)
9. DiffBind: dba.normalize(obj, spikein = TRUE)
10. DESeq2: sizeFactors(dds) <- 1 / scale_factors (inverse convention)
11. edgeR: normFactors via DGEList
12. csaw: apply spike-in factors on the windowCounts SummarizedExperiment (normFactors/normOffsets), not inside windowCounts itself
13. Internal-control sanity check: verify blacklist + housekeeping signal stability post-scaling
14. Differential testing: run DESeq2 / edgeR / csaw with spike-in-based size factors
15. Document: spike-in organism + amount (cells or chromatin mass) + protocol, alignment %, mapq filter, scaling formula, internal-control validation

Tips

• Spike-in is required for global-shift biology. HDACi, BETi, EZH2i, dosage, target knockdown all confound reads-in-peaks normalization.
• Apply at the read level, never to peak counts. A common implementation error in published spike-in ChIP.
• DESeq2 / edgeR sizeFactors use inverse convention. sizeFactors(dds) <- 1 / scale_factors, not direct.
• Pass --scaleFactor alone in bamCoverage for spike-in. deepTools multiplies --scaleFactor by any --normalizeUsing factor, so combining them reintroduces depth normalization.
• Verify titration linearity. Spike-in counts should be 0.5-5% of total reads; outside this range, saturation or noise makes scaling unreliable.
• Filter spike reads to mapq ≥ 30 before counting. Low-mapq reads at low-complexity regions inflate counts spuriously.
• Deduplicate spike reads before counting. PCR duplicates of spike-in don't reflect chromatin amount.
• Use a single enzyme lot for CUT&RUN/CUT&Tag cross-condition comparison. E. coli carryover varies between bacterial batches.
• For high-stakes CUT&Tag cross-condition claims, supplement with deliberate Drosophila spike-in.
• ChIPseqSpikeInFree is diagnostic only, not publication-grade. Re-do experiment with deliberate spike-in if publication is goal.
• The internal-control sanity check is mandatory. Blacklist regions (no biology) should show NO signal change after spike-in scaling. If they do, scaling is broken.

Troubleshooting

Scaling factors all near 1.0

Spike-in not added at fixed amount, OR samples not properly randomized. Re-check Egan 2016 protocol (fixed Drosophila chromatin mass, ~27:1 human:Drosophila genome-copy ratio).

Spike-in fraction <0.1%

Spike-in lost during library prep, OR too few cells used. Scaling unreliable; re-do experiment.

Spike-in fraction >5%

Spike-in too concentrated; saturating. Titration not in linear range; scaling factors unreliable.

Replicate-to-replicate scaling factors vary >2×

Library prep batch effects on spike-in fraction. Re-extract spike reads after deduplication and mapq filter; if still variable, library prep needs improvement.

Results sign-flipped after scaling

Inverse convention bug. sizeFactors(dds) <- 1 / scale_factors, not sizeFactors(dds) <- scale_factors.

Internal-control regions shift after scaling

Scaling broken. Common causes: 1. Scaling applied to peak counts -> fix application layer 2. Spike-in reads not dedup'd -> re-extract from filtered BAM 3. mapq filter too loose -> use -q 30 4. Spike-in saturated -> verify titration linearity

Cross-condition comparison still shows no effect

If spike-in is properly applied and internal controls are stable, the biology may not have the global shift. Consider: 1. Mark biology; H3K27ac may shift less than H3K27me3 with EZH2i 2. Time-course; acute (1 hr) vs chronic (24 hr) differs 3. Inhibitor concentration; sub-saturating doesn't produce global shift

ChIPseqSpikeInFree predicts shift but spike-in says no

Trust spike-in measurement. ChIPseqSpikeInFree is a distribution-shape heuristic; it may misinterpret biology as a global shift.

Related Skills

• chip-seq/peak-calling - Peak calling upstream
• chip-seq/chipseq-qc - Spike-in fraction QC + library complexity
• chip-seq/differential-binding - Apply spike-in scaling via DiffBind / DESeq2 / csaw
• chip-seq/cut-and-run-tag - E. coli spike-in carryover specifics
• chip-seq/super-enhancers - SE calling requires spike-in for cross-condition
• chip-seq/chipseq-visualization - Spike-in-scaled bigWig generation
• alignment-files/sam-bam-basics - Combined-genome alignment
• differential-expression/deseq2-basics - DESeq2 sizeFactors conventions