geo-connector

Search and Cross-Reference NCBI GEO with ENCODE

When to Use

• User wants to find complementary datasets in NCBI GEO to supplement ENCODE data
• User asks about "GEO", "Gene Expression Omnibus", "supplementary data", or "find related datasets"
• User needs to cross-reference ENCODE experiments with GEO series for additional replicates or conditions
• User wants to link ENCODE accessions to GEO/SRA identifiers for data sharing or citation
• Example queries: "find GEO datasets for pancreatic islet RNA-seq", "link this ENCODE experiment to GEO", "search GEO for complementary ATAC-seq data"

Query the Gene Expression Omnibus programmatically to find complementary datasets, cross-reference ENCODE experiments, and download metadata.

Scientific Rationale

The question: "What additional expression or epigenomic datasets exist in GEO that complement my ENCODE analysis?"

GEO hosts >200,000 series across all organisms and assay types. Many ENCODE experiments are deposited in GEO as secondary archives (ENCODE Portal is primary). GEO also contains vast amounts of non-ENCODE data — disease cohorts, perturbation experiments, time courses — that complement ENCODE's reference epigenomes.

GEO ↔ ENCODE Relationship

• ENCODE processed data is deposited at GEO as standard GSE submissions
• Raw sequencing data goes to SRA (linked from both GEO and ENCODE)
• The ENCODE Portal is canonical; GEO is secondary archive
• GEO accessions are stored in ENCODE's dbxrefs field as GEO:GSExxxxx
• NCBI maintains a dedicated ENCODE listing: https://www.ncbi.nlm.nih.gov/geo/encode/

GEO Entity Hierarchy

Series (GSE) — An experiment/study
  ├── Sample (GSM) — Individual measurements
  │     ├── references → Platform (GPL)
  │     ├── has → Supplementary files (raw data)
  │     └── has → Data table (normalized values)
  │
  └── curated into → DataSet (GDS) [not all GSE get curated]
        └── generates → Profiles (gene-level summaries)

Step 1: Find GEO Accessions for ENCODE Experiments

From ENCODE → GEO

ENCODE experiments may have GEO cross-references in their metadata. After tracking an experiment:

encode_track_experiment(accession="ENCSR...")

Check the experiment's dbxrefs field for GEO:GSExxxxx entries. If found, link it:

encode_link_reference(
    experiment_accession="ENCSR...",
    reference_type="geo_accession",
    reference_id="GSE12345"
)

From GEO → ENCODE

Search GEO for ENCODE-deposited data:

# Via NCBI E-utilities
curl "https://eutils.ncbi.nlm.nih.gov/entrez/eutils/esearch.fcgi?db=gds&term=ENCODE[KEYWORD]+AND+gse[ETYP]&retmax=100&usehistory=y&tool=encode_mcp&email=YOUR_EMAIL"

Step 2: Search GEO for Complementary Datasets

E-utilities Search Syntax

Base URL: https://eutils.ncbi.nlm.nih.gov/entrez/eutils/esearch.fcgi

Required parameters: db=gds, term=QUERY, tool=encode_mcp, email=YOUR_EMAIL

Rate limit: 3 req/sec without API key, 10 req/sec with key. Get a key at https://www.ncbi.nlm.nih.gov/account/

Search Field Qualifiers

Qualifier	Purpose	Example
[ETYP]	Entry type	gse[ETYP], gds[ETYP]
[ORGN]	Organism	"Homo sapiens"[ORGN]
[PDAT]	Publication date	2024[PDAT]
[ACCN]	Accession	GPL96[ACCN]
[suppFile]	Supplementary file type	bed[suppFile], bw[suppFile]

Example Searches

# Human pancreas ATAC-seq datasets with BED files
curl "https://eutils.ncbi.nlm.nih.gov/entrez/eutils/esearch.fcgi?db=gds&term=pancreas+AND+ATAC-seq+AND+%22Homo+sapiens%22[ORGN]+AND+gse[ETYP]+AND+bed[suppFile]&retmax=50&tool=encode_mcp&email=YOUR_EMAIL"

# ChIP-seq datasets from a specific year
curl "https://eutils.ncbi.nlm.nih.gov/entrez/eutils/esearch.fcgi?db=gds&term=ChIP-seq+AND+H3K27ac+AND+gse[ETYP]+AND+2024[PDAT]&retmax=50&tool=encode_mcp&email=YOUR_EMAIL"

# Datasets associated with a PubMed ID
curl "https://eutils.ncbi.nlm.nih.gov/entrez/eutils/elink.fcgi?dbfrom=pubmed&db=gds&id=PMID&tool=encode_mcp&email=YOUR_EMAIL"

Step 3: Retrieve GEO Metadata

Get Summary for GEO Records

# Step 1: Search (returns UIDs, NOT accessions)
curl "https://eutils.ncbi.nlm.nih.gov/entrez/eutils/esearch.fcgi?db=gds&term=GSE12345[ACCN]&tool=encode_mcp&email=YOUR_EMAIL"

# Step 2: Get summary (use UID from step 1)
curl "https://eutils.ncbi.nlm.nih.gov/entrez/eutils/esummary.fcgi?db=gds&id=UID&version=2.0&tool=encode_mcp&email=YOUR_EMAIL"

Direct Record Access (acc.cgi)

# Get full SOFT-format record
curl "https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE12345&targ=self&view=full&form=text"

# Get XML (MINiML) format
curl "https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE12345&targ=self&view=full&form=xml"

# Get all sample metadata for a series
curl "https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE12345&targ=gsm&view=brief&form=text"

Step 4: Download GEO Data Files

FTP Directory Convention

GEO uses a "nnn" directory pattern: replace last 3 digits with "nnn".

Accession	FTP Path
GSE12345	ftp://ftp.ncbi.nlm.nih.gov/geo/series/GSE12nnn/GSE12345/
GSM575	ftp://ftp.ncbi.nlm.nih.gov/geo/samples/GSMnnn/GSM575/

Key Download Paths

Content	Path Under Series Directory
Series matrix (expression table)	matrix/GSE12345_series_matrix.txt.gz
SOFT metadata	soft/GSE12345_family.soft.gz
MINiML (XML)	miniml/GSE12345_family.xml.tgz
All supplementary files	suppl/GSE12345_RAW.tar
Individual supplementary	suppl/FILENAME.gz

Download Commands

# Download series matrix (fastest for expression data)
wget "https://ftp.ncbi.nlm.nih.gov/geo/series/GSE12nnn/GSE12345/matrix/GSE12345_series_matrix.txt.gz"

# Download all supplementary files
wget "https://ftp.ncbi.nlm.nih.gov/geo/series/GSE12nnn/GSE12345/suppl/GSE12345_RAW.tar"

Format Selection Guide

Use Case	Format	Speed
Expression matrix analysis	Series matrix	Fastest (10-100x vs SOFT)
Full metadata extraction	SOFT	Complete but slow
XML processing	MINiML	Good for programmatic parsing
Peak/BED files	Supplementary	Direct download
Raw sequencing reads	SRA (not GEO)	Use SRA Toolkit

Step 5: Cross-Reference Workflow

ENCODE + GEO Integration Pattern

1. Find ENCODE experiments of interest:
   encode_search_experiments(assay_title="total RNA-seq", organ="pancreas")

2. For each experiment, check for GEO accession:
   encode_get_experiment(accession="ENCSR...")
   → Look in dbxrefs for "GEO:GSExxxxx"

3. If GEO accession found, link it:
   encode_link_reference(
       experiment_accession="ENCSR...",
       reference_type="geo_accession",
       reference_id="GSE12345"
   )

4. Search GEO for complementary non-ENCODE datasets:
   E-utils search for same tissue + different assay or condition

5. Download GEO metadata for comparison:
   acc.cgi or E-utils esummary

6. Log the cross-reference:
   encode_log_derived_file(
       file_path="/path/to/comparison.tsv",
       source_accessions=["ENCSR...", "GSE12345"],
       description="ENCODE-GEO cross-tissue comparison"
   )

Finding SRA Accessions from GEO

For sequencing data, raw reads are in SRA, not GEO:

# Link GEO to SRA
curl "https://eutils.ncbi.nlm.nih.gov/entrez/eutils/elink.fcgi?dbfrom=gds&db=sra&id=GDS_UID&tool=encode_mcp&email=YOUR_EMAIL"

Python alternative using pysradb:

from pysradb.search import SraSearch
# Convert GSE to SRP
pysradb gse-to-srp GSE12345
# Get all SRR run accessions
pysradb gsm-to-srr GSM12345

Pitfalls and Caveats

1. E-utils return UIDs, not accessions: GEO search returns numeric UIDs. You must call ESummary to get the actual GSE/GDS accession numbers.
2. Not all ENCODE experiments have GEO accessions: The dbxrefs field may be empty. ENCODE Portal is always the canonical source.
3. GEO search max is 10,000 per call: Use usehistory=y and paginate with retstart for large result sets.
4. Series matrix may be split: Large studies can produce multiple matrix files (*_series_matrix-1.txt.gz, *_series_matrix-2.txt.gz).
5. SOFT format discontinued for new submissions (early 2024): Existing records are still downloadable, but new submissions use different formats.
6. Rate limits are strict: Without API key, max 3 requests/second. Include tool= and email= parameters on every request.
7. GEO metadata quality varies: User-submitted metadata may be inconsistent. Always verify organism, platform, and sample characteristics.

Walkthrough: Finding Complementary GEO Datasets for ENCODE Experiments

Goal: Identify Gene Expression Omnibus (GEO) datasets that complement ENCODE epigenomic experiments, enabling integrative analysis of gene expression with regulatory elements. Context: ENCODE provides epigenomic maps (ChIP-seq, ATAC-seq), while GEO hosts vast RNA-seq expression datasets. Combining them links regulatory elements to transcriptional output.

Step 1: Identify the ENCODE experiment to complement

encode_get_experiment(accession="ENCSR123HEP")

Expected output:

{
  "accession": "ENCSR123HEP",
  "assay_title": "Histone ChIP-seq",
  "target": "H3K27ac",
  "biosample_summary": "HepG2",
  "organism": "Homo sapiens",
  "status": "released"
}

Interpretation: This is H3K27ac ChIP-seq in HepG2 (liver cancer cell line). We need matching RNA-seq data from the same cell line.

Step 2: Search GEO for complementary RNA-seq

Using NCBI E-utilities (via skill guidance):

GET https://eutils.ncbi.nlm.nih.gov/entrez/eutils/esearch.fcgi?db=gds&term=HepG2[cell+line]+AND+RNA-seq[strategy]+AND+Homo+sapiens[organism]&retmax=10

Expected response:

{
  "esearchresult": {
    "count": "142",
    "idlist": ["200156789", "200145678", "200134567"]
  }
}

Step 3: Link ENCODE experiment to GEO dataset

encode_link_reference(accession="ENCSR123HEP", reference_type="geo", reference_id="GSE156789", notes="HepG2 RNA-seq for enhancer-expression integration")

Expected output:

{
  "status": "linked",
  "accession": "ENCSR123HEP",
  "reference_type": "geo",
  "reference_id": "GSE156789"
}

Step 4: Check for ENCODE RNA-seq in same biosample

encode_search_experiments(assay_title="total RNA-seq", biosample_term_name="HepG2", organism="Homo sapiens")

Expected output:

{
  "total": 15,
  "results": [
    {"accession": "ENCSR456RNA", "assay_title": "RNA-seq", "biosample_summary": "HepG2", "status": "released"}
  ]
}

Interpretation: ENCODE already has 15 HepG2 RNA-seq experiments! Always check ENCODE first before going to GEO. GEO becomes essential when ENCODE lacks expression data for your specific biosample or experimental condition (e.g., drug treatment, knockdown).

Integration with downstream skills

• GEO expression data feeds into → peak-annotation for enhancer target gene validation
• GEO dataset accessions integrate with → cite-encode for proper data attribution
• GEO RNA-seq complements → gtex-expression for cell-line-specific expression
• GEO metadata links to → data-provenance for full analysis audit trail

Code Examples

1. Link an ENCODE experiment to GEO

encode_link_reference(
  accession="ENCSR000AKA",
  reference_type="geo_accession",
  reference_id="GSE76079",
  notes="Complementary RNA-seq from same lab"
)

Expected output:

{
  "status": "linked",
  "accession": "ENCSR000AKA",
  "reference_type": "geo_accession",
  "reference_id": "GSE76079"
}

2. View all references for a tracked experiment

encode_get_references(accession="ENCSR000AKA")

Expected output:

{
  "accession": "ENCSR000AKA",
  "references": [
    {"type": "geo_accession", "id": "GSE76079", "notes": "Complementary RNA-seq"},
    {"type": "pmid", "id": "27429435", "notes": "Primary publication"}
  ]
}

Integration

This skill produces...	Feed into...	Using tool/skill
GEO accession (GSE/GSM)	Cross-reference link	encode_link_reference(reference_type="geo_accession")
Supplementary expression data	Differential expression	integrative-analysis skill
Complementary replicates from GEO	Expanded sample size	download-encode + batch-analysis
SRA run accessions	Raw data download	bioinformatics-installer (sra-tools)
GEO metadata	Publication cross-reference	cite-encode skill

Related Skills

Skill	When to Use Instead/Additionally
cross-reference	General external reference linking (PubMed, DOI, GEO, NCT)
data-provenance	Logging derived files from ENCODE+GEO combined analyses
search-encode	Finding ENCODE experiments (primary source)
download-encode	Downloading ENCODE files (preferred over GEO for ENCODE data)
track-experiments	Local experiment tracking with GEO cross-references
cite-encode	Getting citations for ENCODE experiments found via GEO
ucsc-browser	Querying aggregated ENCODE tracks at UCSC
publication-trust	Verify literature claims backing analytical decisions

Presenting Results

• Present GEO links as: ENCODE_accession | GEO_accession | title | organism | samples. Suggest: "Would you like to link these GEO accessions to the tracked experiments?"

For the request: "$ARGUMENTS"