broadinstitute

Scaffold a new gbl qc validity check from a plain-language description — a CheckState variant, registry entry, and unit-test fixture — so a domain expert can add a check by reviewing a diff instead of writing Rust from scratch.

Evaluate a gene's tolerance to loss-of-function and missense variation using gnomAD constraint metrics.

Run federation validity checks on a sites-only submission and interpret the QC report. Use when verifying that a partner ran proper QC on data they submitted as AC/AN/nhomalt without sharing genotypes.

Guide through ACMG-style clinical variant interpretation using gnomAD frequency, constraint, pext, and in silico data.

First-run setup for jpx. Installs prerequisites, downloads data, and launches a notebook. Use when someone says "help me get started", "set up this repo", "how do I run this", or opens the repo for the first time.

Add a new annotation source to the JUMP pipeline. Use when integrating external databases like ToxCast, BindingDB, or other compound annotation sources. Guides through notebook creation, workflow.py task wiring, and SQL schema.

Compose a new marimo notebook in jpx by reusing @app.function helpers from the existing catalog (notebooks/nb00_*.py onward) to answer a JUMP Cell Painting question - e.g. "compare these distance metrics", "show consistency for these targets", "plot activity across datasets", "visualize UMAPs colored by X". Trigger whenever the user asks for a notebook, analysis, figure, or vignette that touches JUMP profiles, copairs results, mAP values, target consistency, compound metadata, or morphological activity - even if they don't say "marimo". Use this instead of writing standalone scripts. Also trigger when the user says "write me a notebook that..." or asks to build on existing catalog notebooks.

Work with copairs and copairs-runner for phenotypic profiling analysis. Use when discussing mAP metrics, activity/consistency analysis, configuring copairs-runner YAML configs, reviewing copairs results, or strategizing Cell Painting data analysis.

Compose a new ggsql query in the jx repo to answer a SQL-shaped question against the canonical JUMP metadata DuckDB (`queries/data/jump_metadata.duckdb`) — plate/well/perturbation/compound demographics, source breakdowns, perturbation-type counts, joins across the metadata schema, anything visualizable as a single chart from one SELECT. Trigger when the user asks for a count, distribution, breakdown, summary, or "show me X across Y" question that maps cleanly to SQL + a chart, and the answer fits in one screen. Use this instead of writing a marimo notebook when no Python (image fetching, AnnData profiles, broad-babel translation, copairs computation) is needed. The companion skill `vignette-catalog-compose-notebook` covers the marimo path; cross-check before composing.

Use this skill for gene set provenance, source datasets, gene membership, overlap searches, Pigean phenotype associations, graph neighborhoods, and evidence-backed interpretation from the gene set MCP server.

Find sequencing datasets (NCBI SRA / EMBL-EBI ENA runs) by phenotype or disease, focused on aligned BAM/CRAM reads. Use when the user wants to locate sequencing data, aligned reads, BAM/CRAM files, FASTQ runs, or datasets for a disease, phenotype, or MONDO id.

Manage the gnomad-lr distributed pool for loading long-read VCF data into ClickHouse. Use for creating pools, deploying binaries, submitting jobs, diagnosing failures, and tuning performance.

Interpret variant pathogenicity from gnomAD CSV exports using ACMG/AMP guidelines. Use when analyzing genetic variants, assessing pathogenicity, or interpreting gnomAD data.