Human genome browser¶

Open the browser to explore precomputed GRCh38 SNV delta scores alongside the scoring annotation’s genes, exons and strand. The browser uses the interaction style of the Shorkie yeast browser.

The bundled review dataset contains real scored loci, including a minus-strand gene and alternate-contig reference mismatches. It is explicitly labelled as a subset. Genome-wide publication requires the prepared data on the configured institutional HTTPS host. Complete r10 is the default model; r13 remains a frozen preview with coverage gaps until its final audit passes.

Navigate and inspect¶

Enter a gene symbol, a 1-based inclusive region such as chr1:925851-926130, or an exact allele such as chr1:69091:A>G. Use the chromosome overview, minimap, gene navigation, zoom buttons, or drag to move through the genome. Shift-drag marks a region of interest. The share button captures the dataset snapshot and every view setting. Shared links also pin the immutable manifest URL, so changing the default catalog does not redirect an existing shared view to a different snapshot.

Track settings control order, height, density, score threshold, ALT and gene filters. Six themes and mobile track controls follow the existing yeast browser. The keyboard-accessible variant selector provides the same exact values as clicking the tracks. Sequence search accepts DNA IUPAC symbols and both strands, with visible-region, gene, chromosome and whole-reference scopes. Broad searches require the corresponding static sequence indexes; an unavailable index is reported explicitly.

Interpret the measurements¶

DS_AG / DS_AL: acceptor gain / loss variant delta scores.
DS_DG / DS_DL: donor gain / loss variant delta scores.
DP: original signed offset. The affected genomic coordinate is VCF POS + DP on both gene strands. A zero DS does not establish an affected site, even though its original DP is preserved in the inspector.
ALT heatmap: maximum delta across the displayed gene annotations for each alternate base. Inspect an allele for every original annotation.
r13 coverage: accepted source occurrences, including absent predictions. Pending work, absent predictions, measured zero and positions outside the source collection are distinct states.
r13 − r10: difference in maximum delta for a matched allele/gene. Conflicting annotations, partial preview coverage and REF mismatches are excluded from ordinary comparisons.

r10 and r13 are individual model seeds, with 10,000 nt flanking size, event search distance 50 and M1 masking. They are not a five-model ensemble. The original masked SpliceAI annotation has two-decimal precision and its gene set differs from the MANE scoring annotation. The browser preserves these differences; seed disagreement is not calibrated uncertainty.

At close zoom, exact OpenSpliceAI values retain all five decimal places. Wide views show fixed-bin annotation maxima, never fabricated exact alleles. Summary counts refer to source occurrences or gene annotations and are not labelled unique variants. The original collection does not contain every possible SNV at every reference base. The methods panel reports reference, annotation, model and snapshot hashes.

Export results¶

PNG and vector SVG figures use the same renderer as the tracks. Exact CSV contains every original annotation, signed offsets, affected positions, reference-match flags and source identities. Missing values remain blank; measured zeros remain numeric zero. Summary CSV is a separate export of bin statistics. CSV includes explicitly named 0-based half-open and 1-based coordinate columns. Exports require a successfully loaded, stable view.

Run locally¶

The browser has no model inference dependency. Install Node.js 22.12 or later:

cd docs/genome-browser
npm ci
npm test
npm run dev

Open http://127.0.0.1:4173/OpenSpliceAI/genome/. To use an independently prepared snapshot, start its range server from the repository root:

python -m tools.genome_browser serve results/genome_browser/full

Then open the browser with ?manifest=http://127.0.0.1:8765/manifest.json. The production documentation build runs npm run build and copies the application under genome/.

Prepare and publish genome-wide data¶

Preparation reads accepted, content-hashed VCF audit manifests without changing the scoring files or jobs. It runs on CPUs and creates packed, independently compressed blocks, avoiding millions of tile files. Shards resume only after their configuration and stored content hashes agree.

python -m tools.genome_browser prepare CONFIG.json SNAPSHOT_DIR
# Optional native suffix-array dependency, in an isolated environment:
pip install -r tools/genome_browser/requirements.txt
OMP_NUM_THREADS=2 python -m tools.genome_browser reference CONFIG.json SNAPSHOT_DIR --search
python -m tools.genome_browser finalize CONFIG.json SNAPSHOT_DIR
python -m tools.genome_browser attach-reference SNAPSHOT_DIR
python -m tools.genome_browser verify SNAPSHOT_DIR

The institutional data host must provide HTTPS, HTTP 206 byte ranges, cross-origin access for https://khchao.com and exposed Content-Range. HTTP content compression is disabled for indexed files; blocks are already compressed internally. Upload immutable data files first, verify them, and publish the manifest/catalog last. Set docs/genome-browser/public/settings.json to the public catalog URL. Validate the deployed endpoint with:

python -m tools.genome_browser probe https://HOST/SNAPSHOT/manifest.json

The repository’s tools/genome_browser/README.md documents configuration, resumption, expected resources and publication. FORMAT.md describes the lossless binary format; hosting/ provides Apache and Nginx examples. The small application belongs on GitHub Pages; the large data stay on the institutional host. No paid hosting or GPU inference is needed for the browser. The bundled review subset authenticates small same-origin files before slicing them, which accommodates GitHub Pages recompression. This exception is limited to files at most 1 MiB; genome-wide data always require unchanged range bytes.