tooluniverse-epigenomics-chromatin

# Epigenomics and Chromatin Accessibility Research

## NOT for (use other skills instead)

- Methylation array data processing (CpG beta values, differential methylation) -> Use `tooluniverse-epigenomics`
- RNA-seq differential expression -> Use `tooluniverse-rnaseq-deseq2`
- GWAS variant interpretation -> Use `tooluniverse-gwas-snp-interpretation`
- Variant functional annotation from VCF -> Use `tooluniverse-variant-analysis`

---

## Reasoning: Classify the Question First

Before calling any tool, identify which question type you're answering. Each maps to a different tool set.

**(a) Which regulatory elements exist at a locus?**
Use UCSC_get_encode_cCREs (region-based) or SCREEN_get_regulatory_elements (gene-based). Then check ENCODE_get_chromatin_state for ChromHMM annotation and ENCODE_search_chromatin_accessibility for ATAC-seq evidence.

**(b) Which TFs bind there?**
Use ReMap_get_transcription_factor_binding for ChIP-seq experiments. Use jaspar_search_matrices to retrieve binding motifs and check whether the sequence disrupts a known motif.

**(c) How does a variant affect regulation?**
Use RegulomeDB_query_variant for a scored summary. Then build multi-layer evidence: UCSC_get_encode_cCREs (is the variant in a cCRE?), GTEx_get_single_tissue_eqtls (is it an eQTL?), jaspar_search_matrices (does it disrupt a TF motif?). No single layer is sufficient — see the variant reasoning section below.

**(d) What genes are regulated by an element?**
Use GTEx_get_single_tissue_eqtls or GTEx_query_eqtl to find genes whose expression is associated with variants in the element. Use SCREEN_get_regulatory_elements with element_type="PLS"/"pELS"/"dELS" to classify element-to-promoter relationships.

---

## Reasoning: Histone Marks

Use histone mark identity to guide tool queries and interpret results before fetching data.

- **H3K4me3** = active promoter. If present without H3K27ac, promoter may be active but not hyperacetylated.
- **H3K27ac** = active enhancer or promoter. Strong signal = regulatory element is on.
- **H3K4me1** = poised or active enhancer. Needs H3K27ac to confirm activity; H3K4me1 alone = poised.
- **H3K27me3** = Polycomb repression. Gene is silenced by PRC2.
- **H3K9me3** = constitutive heterochromatin. Region is structurally silenced.
- **H3K36me3** = transcribed gene body. Confirms active elongation.

**Bivalent promoter logic**: If you observe H3K4me3 + H3K27me3 together at the same locus, the promoter is bivalent — poised but not active. This is common in stem cells and developmentally regulated genes. Do not report such genes as "actively transcribed." Use GTEx_get_expression_summary to check if the gene is actually expressed in the tissue of interest.

**Inference rule**: If a user asks about a mark you haven't queried yet, ask: does the mark you *have* found already answer the question? H3K4me3 in a region predicts active transcription; you may not need to also query H3K36me3 unless confirming elongation specifically.

---

## Reasoning: eQTL Interpretation

An eQTL means variant X is statistically associated with expression of gene Y in tissue T. Before reporting eQTL results, apply this chain of reasoning:

1. **Association ≠ causation.** The variant may be in LD with the causal variant. Report effect size (NES) and p-value, not causality.
2. **Check tissue specificity.** Use GTEx_get_multi_tissue_eqtls to see whether the effect is shared across tissues (m-value near 1.0 in many tissues) or tissue-specific (m-value near 1.0 in only one tissue). Tissue-specific eQTLs are stronger candidates for cell-type-specific regulation.
3. **Cross-reference with chromatin.** Is the eQTL variant inside a cCRE? Use UCSC_get_encode_cCREs on the variant's coordinates. If yes, the variant likely acts through a regulatory element.
4. **Check TF motif disruption.** Use jaspar_search_matrices to find motifs overlapping the eQTL locus. If the variant alleles differ in motif score, it is a candidate causal variant.
5. **Effect direction matters.** Positive NES = reference allele increases expression. Negative NES = alternative allele decreases expression.

---

## Reasoning: Variant Regulatory Impact

To assess a non-coding variant's regulatory impact, build evidence from multiple independent layers. No single layer is sufficient.

**Layer 1 — RegulomeDB score**: High probability (score 1a–2b) means convergent evidence from eQTL + TF binding + DNase. Score 4–7 means weak support. Use as a triage filter.

**Layer 2 — Regulatory element overlap**: Query UCSC_get_encode_cCREs at the variant's coordinates. If the variant falls in a cCRE (especially PLS or pELS), it is in a functional context.

**Layer 3 — eQTL evidence**: Query GTEx_get_single_tissue_eqtls for nearby genes. If the variant is a significant eQTL, the association supports regulatory function.

**Layer 4 — TFBS disruption**: Query jaspar_search_matrices for TFs with motifs at the locus. If the variant changes a high-information-content position in a motif, it is a strong functional candidate.

**Synthesis rule**: Report each layer separately. Convergence across 3+ layers = high-confidence regulatory variant. A single layer (e.g., eQTL alone) warrants caution.

---

## Phase 0: Disambiguation

**MyGene_query_genes**: `query` (string). Converts gene symbols to Ensembl IDs and coordinates. Filter results by `symbol == '<GENE>'` — first hit may not match.

**ensembl_lookup_gene**: `gene_id` (Ensembl ID), `species` (REQUIRED, "homo_sapiens"). Returns chr/start/end.

Key format notes:
- GTEx requires versioned GENCODE IDs: `ENSG00000012048.20`
- RegulomeDB takes rsIDs: `rs4994`
- GTEx variant IDs: `chr17_43705621_T_C_b38`
- UCSC cCRE regions: `chrom="chr17", start=7668421, end=7687490`

---

## Phase 1: Histone Modification & ChIP-seq

**ENCODE_search_histone_experiments**: `target` (histone mark), `cell_type` (or `tissue` alias), `biosample_term_name` (most explicit ENCODE ontology name), `limit`.

ENCODE anatomy term notes: "breast" →