Sudo Sequence

Minimal web viewer for finding genomic regions with high between-sample diversity. The pipeline converts multiple aligned sequencing files into fixed-width genomic "patches" and scores each patch by the variance of read counts across samples (days). The Flask app serves the top-scoring patches and lets a simple UI list them.

This project is part of the Cambridge Biohackathon and won first place.

This repo currently operates on BAM alignments (via pysam), not raw FASTQ. If you start from FASTQ, align to a reference and produce sorted BAM files first.

1. to_daru.py loads multiple BAM files and iterates over reads in genomic order.
2. Reads are grouped into fixed-size genomic windows ("patches") based on patch_size.
3. For each patch, a score is computed as the variance of per-sample read counts.
4. Patches are serialized to a compact binary format (daru.daru) with an index (daru.idaru).
5. The Flask app exposes APIs to fetch top patches and per-patch read records.

• src/to_daru.py: offline converter from BAM files to daru binary + index.
• src/daru.py: binary format definitions, read/write helpers.
• src/__init__.py: Flask app exposing APIs and serving the UI.
• src/templates/index.html: minimal Bootstrap UI to list patches.
• setup.sh: creates a virtualenv and installs Python dependencies.
• server.sh: activates the venv and launches the Flask app.

./setup.sh

src/to_daru.py currently hardcodes four BAM paths and writes output to daru.daru and daru.idaru in the repo root.

Update the call at the bottom of src/to_daru.py to point at your BAMs and choose a patch size, then run:

python3 src/to_daru.py

Example expected output files:

• daru.daru
• daru.idaru

./server.sh

Open the app at http://127.0.0.1:5000/.

• GET /api/patches/
  • Returns a JSON map of {patch_byte_offset: score} for the top 200 patches.
• GET /api/patch/<patch_id>
  • Returns a JSON array of read records for the patch at byte offset patch_id.

• Input must be BAM. Convert FASTQ to BAM with your preferred aligner and sort/index as needed.
• Scoring uses variance of read counts per sample; it does not evaluate base-level diversity.
• src/to_daru.py uses query_name for rname in records; if you need reference name, adjust accordingly.
• The UI currently only lists patches; it does not yet plot or visualize reads.

See LICENSE.