OverviewHistoryStatsSecurity
npx skills add ...
Documentation
SKILL.md
npx skills add google-deepmind/science-skills --skill encode-ccres-database
Query the ENCODE Registry of cis-Regulatory Elements (cCREs) via the SCREEN GraphQL API, or make custom queries to the ENCODE Portal REST API for experiments and files (ChIP-seq peaks, etc.). Use when you want to query regulatory annotations or raw experimental data across human cell types.
npx skills add google-deepmind/science-skills --skill encode-ccres-database
This skill allows you to query the ENCODE Registry of cCREs (candidate cis-Regulatory Elements) via the SCREEN GraphQL API. It helps identify functional non-coding DNA elements (like Promoters, Enhancers, and insulators) by analyzing biochemical signatures (DNase, H3K4me3, H3K27ac, CTCF).
uv: Read the uv skill and follow its Setup instructions to ensure
uv is installed and on PATH.cat to read the entire JSON output file
into context, as it can be extremely large. You MUST use jq to efficiently
parse and extract relevant fields.All subcommands write JSON to disk. Always save output in a temporary location
like /tmp/.
Biosamples in ENCODE are often categorized by their data completeness. "Type A" (or high-confidence) biosamples are those that have experimental data for all four core epigenetic markers: DNase, H3K4me3, H3K27ac, and CTCF.
The biosamples and details commands automatically enrich their output with
an is_type_a boolean flag for each biosample.
Example: Finding high-confidence cell types
Do NOT use cat to read the entire JSON output file into context, as it
can be extremely large. Instead, you MUST use jq to efficiently parse and
extract the relevant fields from the JSON file saved by the script. If jq is
not available on the system, write your own Python filtering code (e.g.,
python3 -c "import json...") to extract the necessary data.
For a complete reference of the JSON structure returned by eachmcommand (so you
know which fields to query with jq), read
references/json_output_structure.md.
search: Search cCREs by coordinates, accessions, or epigenetic signals.
nearby-genes: Find nearby genes for given cCRE accessions.
details: Get detailed information and biosample-specific max Z-scores for
a specific cCRE.
biosamples: Get biosample metadata for an assembly.
orthologs: Get orthologous cCREs in another assembly.
linked-genes: Find linked genes via methods like HiC or eQTLs.
gene-expression: Get gene expression (TPM) across all biosamples for a
named gene. Internally resolves the gene symbol to an Ensembl gene ID, then
queries per-biosample RNA-seq quantifications.
entex: Get ENTEx data for a cCRE or genomic region.
gwas: Query genome-wide association studies, SNPs, or enrichment data.
You can supply the --assembly mm10 or --assembly grch38 flag to explicitly
request a specific assembly for most commands. By default, the script targets
grch38 but will automatically fall back to mm10 if no results are found or
if the query fails.
For accessing raw experiments, ChIP-seq peaks, or other datasets that are not
represented as cCREs in SCREEN, use the scripts/encode_portal_api.py script.
It allows custom queries to the ENCODE Portal REST API.
When analyzing .bed or .bigBed files downloaded from ENCODE, standard
bioinformatics tools are highly recommended for finding overlaps (e.g., between
gene promoters and peaks):
bedtools: For fast mathematical operations on genomic intervals.bigBedToBed: For converting binary BigBed files to readable BED
format.pybedtools: A Python wrapper for bedtools.Write custom logic if these tools are not pre-installed.
If you need to make a complex GraphQL query that the script does not support,
read references/graphql_schema.md for a reference of available queries,
arguments, and return fields in the SCREEN GraphQL API.
uv run scripts/screen_api.py biosamples --output /tmp/biosamples.json
# Use jq to filter for Type A biosamples
jq '.data.ccREBiosampleQuery.biosamples[] | select(.is_type_a == true) | .displayname' /tmp/biosamples.json