name	htan
description	Access HTAN (Human Tumor Atlas Network) data — query the portal database, download from Synapse and Gen3/CRDC, query metadata in BigQuery, and search HTAN publications on PubMed.

HTAN Skill

Tools for accessing data from the Human Tumor Atlas Network (HTAN), an NCI Cancer Moonshot initiative constructing 3D atlases of human cancers from precancerous lesions to advanced disease.

First-Time Setup

On first use, check if the htan CLI is available by running htan --version. If it is not installed, guide the user through setup:

Create a venv in the user's project (not in the plugin directory):

uv venv && uv pip install "${CLAUDE_PLUGIN_ROOT}"

Or without uv:

python3 -m venv .venv && source .venv/bin/activate && pip install "${CLAUDE_PLUGIN_ROOT}"

Configure credentials (portal, Synapse, etc.):
```
uv run htan init
```

Allow htan commands — ask the user to add this to their project .claude/settings.json:

{
  "permissions": {
    "allow": [
      "Bash(uv run htan *)"
    ]
  }
}

All htan commands are read-only and safe — credentials are read from local config files, never echoed.

Critical Rules

ALWAYS prefix commands with uv run when a venv exists (e.g., uv run htan query portal tables). NEVER use source .venv/bin/activate && — uv run handles the venv automatically.

NEVER create a virtual environment or install packages inside the plugin cache directory. Venvs go in the user's working directory.

NEVER run htan_setup.py via Bash. It is an interactive wizard that will fail.

CLI Reference

All commands use uv run htan .... NEVER use source .venv/bin/activate. Run any command with --help for full usage.

Portal Database (Recommended Starting Point)

The fastest way to find HTAN files and get download coordinates. Uses the portal's ClickHouse backend.

uv run htan query portal files --organ Breast --assay "scRNA-seq" --limit 20
uv run htan query portal sql "SELECT atlas_name, COUNT(*) as n FROM files GROUP BY atlas_name"
uv run htan query portal tables
uv run htan query portal describe files
uv run htan query portal demographics --atlas "HTAN OHSU"
uv run htan query portal diagnosis --organ Breast --limit 10
uv run htan query portal cases --atlas "HTAN MSK"
uv run htan query portal specimen --organ Colon
uv run htan query portal summary
uv run htan query portal manifest HTA9_1_19512 HTA9_1_19553 --output-dir ./manifests

SQL notes: Array columns (organType, Gender, Race, etc.) require arrayExists(). Use <> instead of !=. LIMIT is auto-applied. See references/clickhouse_portal.md for full schema.

Publications (No Auth Required)

uv run htan pubs search --keyword "spatial transcriptomics" --max-results 5
uv run htan pubs search --author "Sorger PK"
uv run htan pubs fetch 12345678
uv run htan pubs fulltext "tumor microenvironment"

Data Model (No Auth Required)

Query the HTAN Phase 1 data model — 1,071 attributes across 64 manifest components with controlled vocabularies.

uv run htan model components
uv run htan model attributes "scRNA-seq Level 1"
uv run htan model describe "File Format"
uv run htan model valid-values "File Format"
uv run htan model search "barcode"
uv run htan model required "Biospecimen"
uv run htan model deps "scRNA-seq Level 1"
uv run htan model fetch

See references/htan_data_model.md for the full component catalog and identifier patterns.

File Mapping (No Auth Required)

Bridges file IDs to download coordinates using the HTAN portal's DRS mapping (~67,000 files).

uv run htan files lookup HTA9_1_19512
uv run htan files lookup HTA9_1_19512 --format json
uv run htan files update
uv run htan files stats

BigQuery (Requires Google Cloud Credentials)

Deep clinical queries, assay-level metadata (cell counts, library methods, file sizes).

uv run htan query bq tables
uv run htan query bq tables --versioned
uv run htan query bq describe clinical_tier1_demographics
uv run htan query bq sql "SELECT COUNT(*) FROM `isb-cgc-bq.HTAN.clinical_tier1_demographics_current`"
uv run htan query bq query "How many patients with breast cancer?"

See references/bigquery_tables.md for table schemas and query examples.

Downloads

Use native platform CLIs — they're simpler and clearer.

Synapse (open access):

uv run synapse get syn26535909
uv run synapse get syn26535909 --downloadLocation ./data
uv run synapse get -r syn12345678    # recursive folder download

Gen3/CRDC (controlled access):

gen3-client download-single --profile=htan --guid=<guid>

Configuration

uv run htan config check    # Check credential status for all services

Setup

Use uv run htan config check to see what's configured.

Credential storage:

Portal ClickHouse: ~/.config/htan-skill/portal.json (or OS Keychain)
Synapse: SYNAPSE_AUTH_TOKEN env var or ~/.synapseConfig
Gen3: ~/.gen3/credentials.json (requires dbGaP authorization)
BigQuery: Application Default Credentials (gcloud auth application-default login)

No-auth commands (always work, no setup needed):

htan pubs ... — PubMed search
htan model ... — data model queries
htan files ... — file mapping lookups

Access Tier Rules

Data Level / Type	Access	Platform
Level 3-4, Auxiliary, Other	Open	Synapse (`synapseId`)
Level 1-2 sequencing with DRS URI	Controlled	Gen3 (`drs_uri`)
CODEX Level 1, specialized assays (EM, RPPA, slide-seq, mass spec)	Open	Synapse

Workflows

Recommended: Portal → Download (2 steps)

uv run htan query portal files --organ Breast --assay "scRNA-seq" — find files with synapseId and drs_uri
uv run synapse get <synID> (open access) or gen3-client download-single --guid <guid> (controlled)

Alternative: BigQuery → Download (for complex clinical queries)

uv run htan query bq sql "SELECT ..." — find HTAN_Data_File_ID
uv run htan query portal manifest <file_ids> — get download coordinates
Download via uv run synapse get or gen3-client

Platform Data Gaps

Data	Portal	BigQuery
File size	Not available	`File_Size` (INTEGER, bytes) in assay metadata tables
Cell counts	Not available	`Cell_Total` in scRNAseq tables
Library method	Not available	`Library_Construction_Method` in assay tables
Download coordinates	`synapseId`, DRS URI	`entityId` only

Note: File_Size and entityId exist in all BigQuery assay metadata tables (scRNAseq, bulkRNAseq, imaging, scATACseq, electron_microscopy, bulkWES, etc.), not just scRNAseq.

Fallback rule: If a query needs file sizes, cell counts, or assay-level metadata not in the portal, use BigQuery assay metadata tables (e.g., scRNAseq_level3_metadata_current, imaging_level2_metadata_current). Then use portal or htan files lookup for download coordinates.

HTAN Documentation: See references/htan_docs_manual.md for citing HTAN, dbGaP access, data levels, visualization tools.

Atlas Centers: See references/htan_atlases.md for the 14 atlas centers, cancer types, and grant numbers.

Reference Documents

Document	When to Read
`references/clickhouse_portal.md`	Writing portal SQL — schema, array columns, JSON extraction, common mistakes
`references/bigquery_tables.md`	Writing BigQuery SQL — table schemas, naming conventions, example queries
`references/authentication_guide.md`	Setting up credentials for Synapse, Gen3, BigQuery
`references/htan_data_model.md`	Looking up components, controlled vocabularies, identifier patterns
`references/htan_atlases.md`	Atlas centers, cancer types, grant numbers
`references/htan_docs_manual.md`	HTAN Manual site map, citing HTAN, dbGaP access, visualization tools

Security Notes

Credentials: Never log or display tokens, API keys, or credentials in output
SQL safety: All SQL is validated — only SELECT/WITH allowed; write operations blocked
Input validation: DRS URIs, Synapse IDs, table names, and file paths are validated before use
Signed URLs: Gen3 DRS resolution returns sensitive signed URLs — do not log them

name	htan
description	Access HTAN (Human Tumor Atlas Network) data — query the portal database, download from Synapse and Gen3/CRDC, query metadata in BigQuery, and search HTAN publications on PubMed.