Skip to content

09-cross-reference-and-integration.md

Latest commit

 

History

History
164 lines (118 loc) · 7.71 KB

09-cross-reference-and-integration.md

File metadata and controls

164 lines (118 loc) · 7.71 KB

Cross-Database Integration and Multi-Omics Analysis

This vignette demonstrates connecting ENCODE data to external databases and performing multi-omics integration for comprehensive biological insights. All tool outputs shown are from real ENCODE API queries.

Prerequisites: You have tracked experiments locally (Download & Track).

Skills demonstrated: cross-reference, data-provenance, cite-encode, integrative-analysis, regulatory-elements

Scenario

You have accumulated ENCODE experiments across histone ChIP-seq, ATAC-seq, and RNA-seq for human pancreas. Now you need to connect them to GEO, PubMed, and genome browsers while maintaining a complete provenance chain for publication.

Step 1: Link to GEO

You ask Claude: "This ENCODE experiment ENCSR133RZO has GEO data. Link it."

Claude calls: encode_link_reference(experiment_accession="ENCSR133RZO", reference_type="geo_accession", reference_id="GSE187091", description="Original GEO submission for H3K27me3 pancreas ChIP-seq")

{"status": "linked", "experiment_accession": "ENCSR133RZO",
 "reference_type": "geo_accession", "reference_id": "GSE187091"}

Interpretation: GEO may contain supplementary metadata not in ENCODE (donor clinical annotations, custom processing). Reviewers may also require the GEO accession alongside the ENCODE accession.

Step 2: Link to PubMed

You ask Claude: "Link the ENCODE Phase 3 paper to my tracked experiment."

Claude calls: encode_link_reference(experiment_accession="ENCSR133RZO", reference_type="pmid", reference_id="32728249", description="ENCODE Phase 3 integrative analysis (Moore et al., Nature 2020)")

{"status": "linked", "experiment_accession": "ENCSR133RZO",
 "reference_type": "pmid", "reference_id": "32728249"}

Interpretation: PMID 32728249 is now attached to ENCSR133RZO. When you export citations with encode_get_citations, this paper appears automatically. The PMID can also be passed to PubMed MCP tools for abstract retrieval or related articles.

Step 3: Retrieve All External References

You ask Claude: "Show me all external references I've linked."

Claude calls: encode_get_references()

{"references": [
   {"experiment_accession": "ENCSR133RZO", "reference_type": "geo_accession",
    "reference_id": "GSE187091"},
   {"experiment_accession": "ENCSR133RZO", "reference_type": "pmid",
    "reference_id": "32728249"}
 ], "total": 2}

Interpretation: Your library now bridges ENCODE, GEO, and PubMed. Filter by type with encode_get_references(reference_type="pmid"). Supported types: pmid, doi, geo_accession, nct_id (ClinicalTrials.gov), preprint_doi, and other.

Step 4: Annotate with Ensembl Gene Models

You ask Claude: "What Ensembl annotations are relevant for my pancreas peaks?"

Claude recommends:

  • Gene models: GENCODE v44 (GRCh38) -- the annotation set used by ENCODE pipelines. Select the APPRIS principal isoform or MANE Select transcript for TSS consistency.
  • Regulatory build: Ensembl Regulatory Build provides pancreas-relevant promoter, enhancer, CTCF, and open chromatin annotations.
  • TSS window: Use +/- 2 kb from canonical TSS for promoter-proximal classification.

Reference: Frankish et al., Nucleic Acids Research (2021). doi: 10.1093/nar/gkaa1087

Step 5: Visualize in the UCSC Genome Browser

You ask Claude: "How do I visualize my ENCODE data in UCSC Browser?"

Claude recommends three approaches: (1) load ENCODE bigWig files directly as custom tracks via their stable download URLs (e.g., ENCFF387ALH.bigWig for your H3K27me3 signal); (2) register the ENCODE track hub at encodeproject.org/batch_hub/ for automatic metadata labels; (3) layer with ENCODE/SCREEN CREs, PhyloP 100-way conservation, and RepeatMasker for genomic context.

Reference: Nassar et al., Nucleic Acids Research (2023). doi: 10.1093/nar/gkac1072

Step 6: Multi-Omics Integration Framework

You ask Claude: "How do I integrate RNA-seq, ChIP-seq, and ATAC-seq from pancreas?"

Claude uses the integrative-analysis skill to recommend:

  1. Assembly alignment. All experiments must use GRCh38. Restrict to the same biosample type (tissue). Mixing tissue with cell line introduces culture-adapted confounds.
  2. Chromatin state classification. Combine marks to define regulatory states:
    • Active promoter: H3K4me3+ / H3K27ac+ / ATAC+ / RNA high
    • Active enhancer: H3K4me1+ / H3K27ac+ / ATAC+ / H3K4me3-
    • Poised enhancer: H3K4me1+ / H3K27me3+ / ATAC low
    • Repressed: H3K27me3+ / H3K9me3+ / ATAC- / RNA low
  3. Statistical integration. ChromHMM for state segmentation, GREAT for enhancer-gene association, ATAC-RNA correlation at promoters for expression prediction.
  4. Batch awareness. Check signal distributions with MA plots before combining experiments from different labs.

References: Ernst & Kellis, Nature Methods (2012) doi: 10.1038/nmeth.1906; McLean et al., Nature Biotechnology (2010) doi: 10.1038/nbt.1630

Step 7: Generate a Methods Section with Full Provenance

You ask Claude: "Help me write the methods section for my paper."

Claude retrieves tracked experiments, linked references, and derived file records to generate a structured methods block:

Data acquisition. Histone ChIP-seq data for H3K27me3 in human pancreas tissue were obtained from ENCODE (ENCSR133RZO; GEO: GSE187091). Processed peak calls (ENCFF635JIA, bed narrowPeak, GRCh38) and signal tracks (ENCFF387ALH, bigWig) were downloaded with MD5 verification. Data processed by ENCODE ChIP-seq pipeline v2.

Post-processing. Peaks filtered against the ENCODE blacklist v2 (Amemiya et al., 2019) using bedtools subtract. Parameters recorded in provenance record prov_001.

Citation. Data generated by the ENCODE Consortium (Moore et al., 2020; PMID 32728249). Experiment accessions listed in Supplementary Table 1.

Why this matters: Every accession, tool version, and parameter is traceable. The encode_get_provenance tool retrieves the full chain from any derived file back to its ENCODE source. A reviewer can reconstruct your analysis from identifiers alone.

Best Practices

  • Link references early. Add GEO, PubMed, and DOI links when you first track an experiment, not at submission time. Memory fades; metadata does not.
  • Use permanent identifiers. ENCODE accessions, GEO accessions, PMIDs, and DOIs are stable. Avoid linking to URLs that may change.
  • One assembly, one annotation. Never mix GRCh38 peaks with hg19 gene models. LiftOver first and document the chain file version.
  • Record everything. Use encode_log_derived_file for every processing step. The cost is seconds; the value at review time is immeasurable.
  • Export for collaborators. encode_export_data(format="csv") shares your tracked library as a spreadsheet with PMIDs ready for PubMed lookup.

Skills Demonstrated

  • cross-reference -- Linking ENCODE to GEO, PubMed, and external databases
  • data-provenance -- Traceable chain from source data to derived results
  • cite-encode -- Publication-ready citations with proper ENCODE attribution
  • integrative-analysis -- Combining histone, accessibility, and expression data
  • regulatory-elements -- Classifying active, poised, and repressed chromatin states

What's Next

This vignette covers the final step in a typical ENCODE workflow. For the full path: