A web application for running Similarity Ensemble Approach (SEA) searches and viewing results. SEA relates proteins based on the set-wise chemical similarity among their ligands and can be used to rapidly search large compound databases and build cross-target similarity maps.
Provided by the Shoichet Laboratory in the Department of Pharmaceutical Chemistry at the University of California, San Francisco (UCSF).
| Service | Description |
|---|---|
| web | FastAPI backend serving the API and (in production) the built React frontend |
| redis | Message broker and result backend for Celery |
| redisworker | Celery worker processing SEA scoring tasks across 4 priority queues (sea-slot-0 through sea-slot-3) |
| beat_scheduler | Celery beat scheduler for periodic tasks (e.g. daily job directory cleanup) |
| boltzworker-gpu{0-3} | Per-GPU Celery workers running Boltz2 structure predictions; each listens on boltz-gpu{n} (manual) and boltz-gpu{n}-auto (auto-queued from SEA results) queues |
| frontend | Vite dev server with HMR (development only) |
- Docker and Docker Compose
- Access to the SEA data folder and job directory (NFS or local paths)
- NVIDIA GPUs with Docker GPU support for Boltz2 structure prediction
Create a .env file in the project root:
REDIS_HOST=redis
REDIS_PORT=6380
CELERY_BROKER_URL="redis://redis:6380/0"
CELERY_RESULT_BACKEND="redis://redis:6380/1"
MAX_QUEUE_SIZE=50
SEA_DATA_FOLDER="/path/to/sea_data"
SEA_JOB_DIR="/path/to/sea_jobs"
# SEA parameters
TC_CUTOFF=0.29
MEAN_PARA=4.24 * 10 ** -4
STD_COEF=4.49 * 10 ** -3
STD_EXP=0.665
RADIUS=2
FP_SIZE=2048
USE_FEATUREs=False
# Boltz2 structure prediction
BOLTZ_NUM_GPUS=4
BOLTZ_MAX_QUEUED=10
BOLTZ_BIN=/opt/conda/envs/sea25/bin/boltz
BOLTZ_CACHE_DIR=/scratch/sea_boltz
# ALTCHA captcha HMAC secret (generate your own)
ALTCHA_HMAC_SECRET="your-secret-here"Uses Dockerfile.dev with a separate Vite dev server for frontend HMR, source-mounted backend for live reload, and lower worker concurrency.
docker compose -f compose.dev.yml up --build- Frontend (Vite HMR): http://localhost:5173
- Backend API: http://localhost:5055
Uses Dockerfile.prod with a multi-stage build that compiles the React frontend and bundles it into the backend image.
docker compose -f compose.prod.yml up --build -d- Application: http://localhost:5050
conda env create -f environment.yml
conda activate sea25
pip install -r requirements.txtredis-server --port 6380 --daemonize yesuvicorn backend.app.api:app --host 0.0.0.0 --port 5050 --reloadcelery -A backend.app.celery_worker worker --beat -s /tmp/sea-celerybeat-schedule --loglevel=info --concurrency=5 --max-tasks-per-child=10Start one Boltz2 worker per GPU (GPU index passed via queue name convention):
# Repeat for each GPU, substituting the GPU index
CUDA_VISIBLE_DEVICES=0 celery -A backend.app.celery_worker worker \
-Q boltz-gpu0-auto,boltz-gpu0 --concurrency=1 --pool solo --loglevel=infocd frontend
npm install
npm run dev