← Về thư mục

Allele-Specific Binding - Usage Guide

Overview

Identify variants with allelic effects on TF or histone modification binding from heterozygous-variant ChIP-seq. Covers WASP (mandatory reference-bias filter), RASQUAL (joint cis-QTL + bias-aware testing), BaalChIP (Bayesian beta-binomial with copy-number-aware overdispersion), and AlleleSeq (personalized diploid genome). Embeds the three universal pitfalls (reference bias, imprinted loci, X-inactivation, copy-number imbalance) and integrates with downstream caQTL / bQTL / fine-mapping.

Prerequisites

# WASP (mandatory upstream)
git clone https://github.com/bmvdgeijn/WASP.git
# Requires Python 3, pysam, pytables

# Aligner
conda install -c bioconda bowtie2 samtools bcftools

# Variant calling (for sample-specific hetSNPs)
conda install -c bioconda gatk4

BiocManager::install(c('BaalChIP', 'AllelicImbalance'))

# RASQUAL
git clone https://github.com/dg13/rasqual.git
# Build with `make` in repo

# AlleleSeq
# https://github.com/gerstein-lab/AlleleSeq

Quick Start

Tell the agent what to do: - "Apply WASP reference-bias filter to my ChIP-seq BAM before any ASB analysis (mandatory step)" - "Run BaalChIP on TNBC cancer samples with copy-number-aware overdispersion" - "Use RASQUAL for joint cis-QTL + ASB analysis on a population of 50 samples" - "Build a personalized diploid genome with AlleleSeq for a single-sample ASB analysis" - "Filter imprinted loci and X-inactivated chrX before reporting ASB" - "Validate chromBPNet variant effect predictions against ASB measurements"

Example Prompts

WASP pipeline

"Run the WASP mapping pipeline on a bowtie2-aligned ChIP-seq BAM with my sample's heterozygous SNP VCF. Apply the three steps: find intersecting SNPs, re-map swapped reads, filter discordant mappings. Output WASP-filtered BAM (will lose 22-31% of reads)."

BaalChIP for cancer

"I have FOXA1 ChIP-seq in HCC1395 (TNBC). Run BaalChIP with copy-number-aware overdispersion using my ASCAT CNV calls. Output ASB sites with posterior > 0.95."

RASQUAL for population

"I have 50 individuals' CTCF ChIP-seq with phased genotypes. Run RASQUAL for cis-QTL mapping with built-in bias correction (phi parameter). Window 250 kb around each peak."

AlleleSeq personalized genome

"Build a personalized diploid genome from my sample's phased VCF. Align ChIP reads to maternal and paternal copies separately, then merge for binomial ASB test."

Cross-validation with DL

"I have chromBPNet variant predictions and BaalChIP ASB measurements. Compare: variants with |log2_fc| > 1 in chromBPNet should show ASB; variants with no ASB but predicted strong effect may be in extrapolation regime."

Filter universal pitfalls

"Before reporting ASB: (1) verify WASP was applied; (2) filter imprinted loci (H19, IGF2, MEG3, KCNQ1OT1); (3) exclude chrX in female samples; (4) for cancer, exclude CN-altered regions or use BaalChIP with CN file."

What the Agent Will Do

1. Verify hetSNP source: sample-specific genotype VCF (from GATK / DeepVariant) preferred over population SNP panel
2. Apply WASP filter: mandatory; produces unbiased BAM with 22-31% read loss
3. Filter known biology:
4. Imprinted loci (H19, IGF2, etc.) from imprinted-genes BED
5. chrX in female samples (X-inactivation)
6. Copy-number-altered regions (for cancer; or use BaalChIP)
7. Choose ASB method:
8. Single sample, normal: BaalChIP or AlleleSeq
9. Cancer sample with CNV: BaalChIP with CN file
10. Population (>20 samples): RASQUAL for joint cis-QTL
11. Need maximum bias correction: AlleleSeq with phased genome
12. Run test with appropriate parameters
13. Output: per-hetSNP ASB calls with allelic ratio, posterior, FDR
14. Cross-validate with deep-learning predictions (chromBPNet, EnFormer) where applicable
15. Document: WASP version + filter loss, ASB method + parameters, imprinted/X-filter applied, CN integration

Tips

• WASP is mandatory. Skipping it produces reference-allele-skewed false positives.
• Use sample-specific hetSNP VCF, not population panel. Sample-specific variants may not be in 1KG.
• WASP drops 22-31% of reads. Plan sequencing depth accordingly; ASB power requires roughly 2× the depth needed for peak calling.
• For cancer, BaalChIP with CN file is the only rigorous option. CN imbalance changes effective allele dose; methods that ignore it produce false ASB calls.
• Imprinted loci and X-inactivation are the most common artifact sources. Always filter.
• AlleleSeq is conceptually cleanest but expensive. Personalized diploid genome construction + alignment doubles compute cost.
• RASQUAL phi parameter is the per-feature bias estimate; can be cross-checked against WASP filter loss.
• Validate DL variant predictions against measured ASB. chromBPNet predicts in counterfactual; ASB measures in actual sample. Concordance increases confidence.
• ASB ≠ caQTL. ASB is single-sample at hetSNPs; caQTL is population-level association of any variant with chromatin accessibility / binding.

Troubleshooting

Many ASB calls still REF-biased

WASP not applied or sample-specific hetSNPs missing from WASP SNP file. Build WASP table from sample's own genotype VCF.

Most ASB calls at imprinted loci

Imprinted loci not filtered. Apply imprinted-genes BED before reporting.

Most ASB at chrX (female sample)

X-inactivation. Filter chrX or use X-inactivation-aware methods.

Bimodal allelic ratios in cancer

Copy-number imbalance. Use BaalChIP with CN file OR exclude CN-altered regions.

BaalChIP "no hetSNPs in peaks"

hetSNP / peak chromosome naming mismatch. Standardize chr prefixes.

RASQUAL convergence failure

1. Sparse hetSNPs per peak -> require more samples or combine peaks
2. Strong CN imbalance -> use BaalChIP
3. Imputation quality poor -> require r² > 0.8

AlleleSeq "diploid genome too large"

Many large SVs in genome. Use small-variant-only VCF; exclude SV-rich regions.

Low statistical power

ASB needs ≥20 reads per allele to detect a 2:1 ratio at p<0.05. Increase sequencing depth (2× normal ChIP for ASB power) or combine replicates.

Disagreement with chromBPNet predictions

1. ASB sample has low coverage at variant -> boost depth
2. chromBPNet model may be in extrapolation regime -> check ensemble agreement
3. Variant in context the model didn't see -> caution

Related Skills

• chip-seq/peak-calling - Peak calling upstream
• chip-seq/chipseq-qc - QC before ASB
• chip-seq/chip-deep-learning - Cross-validate ASB with DL variant predictions
• chip-seq/peak-annotation - Annotate ASB variants to genes / cCREs
• atac-seq/allele-specific-accessibility - Parallel ATAC ASB workflow
• causal-genomics/fine-mapping - ASB as fine-mapping orthogonal evidence
• variant-calling/variant-annotation - Annotate hetSNPs
• phasing-imputation/haplotype-phasing - Required for AlleleSeq