BRaVa browser

A fast, static browser for Biobank Rare Variant Analysis (BRaVa) consortium gene-level rare coding-variant association results — gene-level meta-analysis of ~1.2M individuals across 10 global biobanks. Modeled on gnomAD / Genebass.

• Landing — Google-style search over genes (symbol / Ensembl ID) and traits.
• Gene page — phenome-wide associations (PheWAS), a cross-ancestry forest plot (β ± 95% CI per ancestry + meta + heterogeneity p), and a sortable results table.
• Phenotype page — canvas Manhattan plot + virtualized results table; click any gene for its cross-ancestry forest.

Filter everything by ancestry (cross-ancestry meta · EUR/AFR/AMR/EAS/SAS/non-EUR), variant mask, MAF cutoff, and test (Burden / SKAT / SKAT-O).

Architecture

The raw data (~8 GB of gzipped TSVs, originally in GCS — see docs/local-notes.md, gitignored, for the bucket path) is too large to load in-browser, so the project is two halves:

1. pipeline/ — a Python (Polars) ETL that transforms the raw TSVs into compact, columnar JSON: one file per gene, one per (phenotype × ancestry), plus small bundled metadata indexes. Output is uploaded to a Cloudflare R2 bucket.
2. app/ — a React + Vite + TypeScript single-page app hosted on GitHub Pages. It bundles the small search indexes and fetches the per-gene / per-phenotype files from R2 over HTTPS.

GitHub Pages (app + meta indexes)
        │ fetch()
        ▼
Cloudflare R2 (brava-browser bucket): {gene,phenotype}/… , v2/variant/…

Develop the app

cd app
npm install
npm run dev        # http://localhost:5173 (uses bundled sample data)

By default the app reads sample data from app/public/data (a few traits + example genes, committed so the app runs offline). Point it at the full dataset on R2 with VITE_DATA_BASE_URL / VITE_VARIANT_BASE_URL (see .github/workflows/deploy.yml for the live values).

Build / refresh the data (pipeline/)

cd pipeline
pip install -r requirements.txt   # polars; needs gsutil authenticated
export BRAVA_RAW_BUCKET=gs://...  # the raw-data bucket (see docs/local-notes.md)

make meta      # gene + phenotype metadata indexes -> ../app/public/data/meta
make sample    # small local dataset for dev
make full      # full dataset -> ./build  (~45 min, sharded to bound memory)
make upload-genes  # push ./build/{gene,phenotype} to the live R2 bucket

make upload-genes requires rclone configured with an R2 S3-compatible remote (rclone config; see the header comment in pipeline/Makefile).

Deploy

1. CORS (once): allow the Pages origin to fetch. Data is hosted on Cloudflare R2, so set the bucket's CORS policy in the dashboard (R2 → brava-browser → Settings → CORS Policy), pasting infra/cors.json. The object-scoped R2 API token cannot edit CORS, so this is dashboard-only.
2. Pages: enable GitHub Pages → Source: GitHub Actions. Pushing to main runs .github/workflows/deploy.yml, which builds the app (with VITE_DATA_BASE_URL / VITE_VARIANT_BASE_URL set to the R2 bucket) and deploys.

Data notes

• Gene-level only (v1). Each (gene, mask, MAF) carries Burden / SKAT / SKAT-O p-values; effect size β and SE come from the inverse-variance-weighted Burden meta, along with the cross-cohort heterogeneity p (Pvalue_het).
• Gene symbols / positions are joined from Ensembl 110 (GRCh38); phenotype names, categories, and binary/quantitative class are parsed from the BRaVa curation repo.
• Summary statistics only — not for clinical use.