Informatics Data Skills

A collection of Claude Code Agent Skills for obtaining metadata and data from public life-science data repositories — plus helpers that turn what they find into pipeline-ready sample sheets, cluster download scripts, and a data-portal submission workbook.

Each skill gives an agent (or a human at the CLI) one consistent way to:

1. Search a repository and fetch metadata for an accession.
2. List and download the associated data files.
3. Produce pipeline inputs — a harmonized metadata table, nf-core sample sheets, quantms/DIA-NN minimal SDRFs, and cluster-ready download / job scripts.
4. Produce a submission workbook — fill the UK DRI FAIR-metadata Excel template for the ADDI / AD Workbench data portal.

Skills

Skill	Repository	Operator
geo	Gene Expression Omnibus	NCBI
ena	European Nucleotide Archive	EMBL-EBI
pride	PRIDE (proteomics identifications)	EMBL-EBI
arrayexpress	ArrayExpress (functional genomics)	EMBL-EBI (in BioStudies)
biostudies	BioStudies	EMBL-EBI
fastq-download-script	FASTQ download-script generator	—
sra	NCBI SRA download job-script generator (sra-tools)	—
addi	ADDI / AD Workbench FAIR-metadata submission workbook	—

One skill = one directory = one data source (or one generator).

What each skill does

• geo — Search GEO DataSets, fetch series/sample metadata, list and download supplementary/matrix/SOFT/MINiML files, and export the SRA Run Selector list. Raw reads are not held in GEO, so the sample-sheet builder resolves the linked SRA/ENA data for FASTQ links.
• ena — The metadata hub for sequencing runs, and the FASTQ source for bulk data. List runs and download links, run advanced metadata searches, fetch raw records, and download FASTQ / submitted files. Other sequencing skills resolve their runs through ENA; single-cell reads are better fetched with sra (see below).
• pride — Fetch proteomics project metadata, list and download data files (RAW, mzIdentML, mzML, mzTab, MGF, SDRF), search projects, write a minimal-valid SDRF sample sheet for quantms/DIA-NN, and generate a download script for the project's data files.
• arrayexpress — Fetch functional-genomics study metadata with MAGE-TAB (IDF/SDRF) awareness, classify and download files, print the experimental design, and build an nf-core sample sheet from the SDRF.
• biostudies — Fetch study metadata, list and download attached files, and search across BioStudies collections (ArrayExpress, BioImage Archive, and standalone submissions).
• fastq-download-script — Turn a sample sheet (or a plain URL list) into a cluster-ready bash download script with a SLURM header.
• sra — Turn an accession list into sequential SLURM job scripts (prefetch → fasterq-dump, plus an optional symlink step that gives 10x reads the _S1_L001_R{1,2}_001.fastq.gz naming cellranger requires) that download SRR accessions from NCBI SRA and extract them to gzipped FASTQ. The recommended route for single-cell reads — only fasterq-dump --include-technical exposes the Chromium read structure reliably — and the route for anything not mirrored to ENA.
• addi — Fill the UK DRI FAIR-metadata Excel template for an ADDI / AD Workbench submission: describe the study (catalogue) and its data tables (dictionaries/fields/lookups), validate against the template's controlled vocabularies and rules, and write a filled .xlsx — optionally seeding the schema from a metadata.tsv. The template is edited in place so its dropdowns, colors and comments are preserved. Writes files only; never uploads. Requires openpyxl + pandas.

Cross-cutting features

• Harmonized metadata table — every repository skill can write a single metadata.tsv that maps each source's differently-named sample annotations into one common schema (sample, species, sex, age, condition, genotype, treatment, tissue, …), one row per sample × replicate, with any extra characteristic promoted to its own column.
• Sample sheets — the format follows the target pipeline: nf-core/scrnaseq or nf-core/rnaseq for sequencing (chosen with --assay), or a quantms/DIA-NN minimal SDRF for proteomics.
• Download & job scripts — generators that emit cluster-ready transfer scripts (direct FASTQ download from ENA/ArrayExpress) or a two-step sra-tools compute job for NCBI SRA (three-step with --cellranger-links). They only write the scripts; they never submit them.

Typical end-to-end routes — split by assay, not ranked overall:

# BULK RNA-seq — ENA direct HTTPS download
ena samplesheet --assay bulk  →  samplesheet.csv  →  fastq-download-script  →  sbatch

# SINGLE-CELL — the sra route, for the 10x read structure
geo runtable  →  SRR_Acc_List.txt
sra job-scripts --cellranger-links --read-map 3,4  →  3 scripts  →  sbatch (in order)
geo samplesheet --assay scrna --fastq-dir ./fastq --fastq-naming cellranger
    →  samplesheet.csv  →  nf-core/scrnaseq (aligner cellranger)

# PROTEOMICS — a different domain; no reads, no R1/R2
pride download-script  →  sbatch;   pride samplesheet  →  .sdrf.tsv

Installing the skills

Claude Code discovers skills under ~/.claude/skills. Symlink each skill directory there so it stays in sync with this repository:

# Symlink every skill in this repository into ~/.claude/skills
mkdir -p ~/.claude/skills
REPO="$(pwd)"                     # run from the repository root
for d in geo ena pride arrayexpress biostudies fastq-download-script sra addi; do
    ln -s "$REPO/$d" ~/.claude/skills/"$d"
done

To add a single skill, symlink just that directory:

ln -s "$(pwd)/geo" ~/.claude/skills/geo

Because these are symlinks, git pull in this repository updates the installed skills automatically. Remove a skill with rm ~/.claude/skills/<name> (this deletes only the link, not the repository).

Design

See DESIGN.md for the architecture, conventions, and rationale behind every skill.

License

MIT © 2026 UK Dementia Research Institute