OpenKinetics Data

Data portal for CatLog-derived enzyme kinetics data served by OpenKinetics.

This project backs data.openkinetics.org. It is separate from the predictor application, but is designed to deploy on the same server and match the predictor site's general visual style.

Current Sample Data

The current sample artifact is:

data/sample/openkinetics_demo_100.json

It contains 100 CatLog-derived datapoints with UniProt sequences and PubChem substrate structures joined in. The sample is not the full CatLog source snapshot.

Attribution

This data resource is built from CatLog, developed by the Chowdhury Lab and collaborators. The full CatLog publication is coming soon. For now, please cite:

Sajeevan et al., Robust Prediction of Enzyme Variant Kinetics with RealKcat, bioRxiv 2025, DOI: https://doi.org/10.1101/2025.02.10.637555

CatLog static browser: https://chowdhurylab.github.io/tools/catlog-static/

Local Backend

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
python backend/manage.py migrate
python scripts/build_release_files.py
python backend/manage.py import_release
python backend/manage.py runserver 8001

Local Frontend

cd frontend
npm install
npm run dev

The frontend expects the Django API at http://localhost:8001 unless VITE_API_BASE_URL is set.

Docker Deployment

Production is expected to run from:

/home/saleh/openkinetics-data

The Docker setup builds two images:

• backend: Django + Gunicorn on port 8010 inside the Compose network.
• frontend: Vite build served by Nginx on host port 8082, or the bind address set by OPENKINETICS_FRONTEND_BIND.

Persistent state lives on the server and is bind-mounted into containers:

• ${OPENKINETICS_RUNTIME_HOST_DIR:-./runtime}:/data/runtime stores the SQLite database.
• ${OPENKINETICS_RELEASES_HOST_DIR:-./releases}:/data/releases stores release files and zip downloads.
• /home/saleh/webKinPred/media/sequence_info:/sequence_info:ro exposes existing predictor sequence artifacts without copying them.

runtime/ and releases/ are deployment state. Keep them on the server and out of git. For production, set OPENKINETICS_RELEASES_HOST_DIR to a persistent server directory such as /srv/openkinetics-data/releases if you do not want release outputs under the repo checkout at all.

Create .env from .env.example. On production, set OPENKINETICS_FRONTEND_BIND=10.1.2.12:8082, then run:

mkdir -p runtime releases
docker compose build
docker compose run --rm backend python backend/manage.py migrate
docker compose run --rm backend python scripts/build_release_files.py
docker compose run --rm backend python scripts/build_sequence_artifact_bundles.py
docker compose run --rm backend python backend/manage.py import_release
docker compose up -d

For a full release JSON, the preferred production command is:

scripts/publish_release_from_json.sh /path/to/openkinetics_release.json

The script builds release files under ./releases, generates missing sequence artifacts through the GPU service, imports the release as latest, and restarts the website. If OPENKINETICS_RELEASES_HOST_DIR is set, the script writes there instead. Set GPU_EMBED_SERVICE_URL and GPU_EMBED_SERVICE_TOKEN in the environment or pass --gpu-service-url and --gpu-service-token.

The frontend container serves the React app, proxies /api/, /admin/, and /sequence-artifacts/ to Django, and serves /releases/ from the mounted release directory.

Sequence artifact files are keyed by sequence_id, matching the predictor seqmap.sqlite3 ID where possible and otherwise using sha256(sequence)[:12]. Mounted source arrays live at paths such as:

/sequence_info/esm2_layer_33/residue_vecs/{sequence_id}.npy
/sequence_info/esmc_layer_32/residue_vecs/{sequence_id}.npy
/sequence_info/prot_t5_last/residue_vecs/{sequence_id}.npy
/sequence_info/pseq2sites_scores/{sequence_id}.npy

Per-record artifact downloads are served as ZIP packages containing README.txt, manifest.json, sequence.json, and the raw .npy array. Pseq2Sites downloads also include a readable pseq2sites/scores.json payload with sequence_id, sequence, and per-residue scores when the score vector can be parsed. Release-level embedding and Pseq2Sites bundles include metadata/sequences.jsonl and metadata/artifacts.jsonl so every array is distributed with its amino-acid sequence.