WOMPBAT

A frontend for exploring polygenic score (PGS)–phecode associations produced by researchers at the Colorado Center for Personalized Medicine (CCPM) and the University of Colorado Anschutz Medical Campus.

Production: wompbat.simonwalker.tech

The underlying data and scientific analyses are described in:

Meng Lin et al. (2026). The landscape of polygenic score prediction accuracy and transportability across the phenome in ancestrally diverse populations. medRxiv preprint. Read the paper.

This repository provides the browser interface and helpers that package the research results for display. Refer to the paper for study methods and interpretation, and cite it when using the underlying research.

Explore the data

• Search by PGS ID or trait, or by phecode or phenotype name.
• Switch between continuous and thresholded PGS results.
• Inspect association tables, ancestry-specific effects, forest plots, and case rates.
• Filter for low heterogeneity (I² < 20) and download results as TSV.

The site serves precomputed results: by default, up to 20 FDR-passing associations per query and PGS unit, ranked by mixed-model p-value. The heterogeneity filter applies to those results without filling in additional rows.

Local development

Use Bun and Snakemake. The optional Nix development shell provides both, along with just:

nix develop

Install dependencies and generate the query assets from the committed data cache:

bun install --frozen-lockfile
gunzip -k data/precomputed.json.gz
snakemake --snakefile workflow/Snakefile --cores all static/pgs static/phecode
bun run dev

Decompress the cache once on a fresh checkout; if data/precomputed.json already exists, skip that step. Snakemake consumes this intermediate file when splitting the cache into query assets.

The frontend uses SvelteKit, Svelte 5, TypeScript, UnoCSS, and svelteplot. It fetches static JSON files from static/pgs/ and static/phecode/. Autocomplete reads the committed static/autocomplete.db in the browser using sql.js; bun install copies its WebAssembly runtime into static/. No backend service or credentials are needed to run the frontend.

Build and deploy

bun run build
bun run preview

or just dev

Generate the query assets first, as above. SvelteKit's static adapter writes the site to build/.

Pushes to main trigger the deployment workflow, which expands the committed cache, generates the static query assets, builds the app, and deploys to GitHub Pages. Production is served at wompbat.simonwalker.tech.

Validation

bun run check
bun run test
bunx playwright install chromium
bun run test:integration

The database tests and browser integration suite require data/megatable.db; the browser suite also requires generated query assets. Integration tests compare downloaded TSVs against independent queries of that source database. To check production:

INTEGRATION_BASE_URL=https://wompbat.simonwalker.tech bun run test:integration

See the integration test guide for prerequisites, query counts, comparison rules, and reports.

Data preparation

Frontend development uses the committed data/precomputed.json.gz cache and static/autocomplete.db. Rebuilding them requires the research data locally; the source megatable and data/megatable.db are excluded from Git.

The Snakemake workflow includes helpers to build the SQLite database (build_ccpm_db), precompute browser results (build_precomputed_index), compress the cache (compress_precomputed_index), split query assets (split_precomputed_index), and rebuild autocomplete (build_autocomplete_db). These package existing research outputs for the viewer. Column mappings are documented in docs/columns.md.