addie-models¶

ADMET (absorption, distribution, metabolism, excretion, toxicity) prediction. 31 pretrained ML models covering CYP inhibition/substrate, clearance, half-life, hepatotoxicity, cardiotoxicity (hERG + DICTrank), DILI, Ames, and Tox21 endpoints.

Pre-reqs¶

Docker
CPU works; GPU (any NVIDIA) speeds up batch inference ~10×
~4 GB RAM
~2 GB disk for weights (bundled in the image)

Deploy¶

Via docker-compose:

Uncomment the addie-models block in docker-compose.yml, then:

docker compose up addie-models

Standalone:

# CPU only
docker run -d \
  --name novomcp-addie \
  -p 8033:8033 \
  --restart unless-stopped \
  ghcr.io/novomcp/addie-models:latest

# With GPU
docker run -d \
  --name novomcp-addie \
  --gpus all \
  -p 8033:8033 \
  --restart unless-stopped \
  ghcr.io/novomcp/addie-models:latest

Wire into the engine¶

export ADDIE_MODELS_URL=http://localhost:8033

Verify¶

curl -s http://localhost:8033/health
# {"status":"healthy","models_loaded":31,"gpu_available":true|false}

curl -s -X POST http://localhost:8033/predict \
  -H 'Content-Type: application/json' \
  -d '{"smiles":"CC(=O)Oc1ccccc1C(=O)O"}'

Tools that light up¶

predict_admet, 31 endpoints per molecule
get_molecule_profile, fills the ADMET section (falls back to properties-only without this service)
screen_library, batch ADMET across a compound list
batch_profile, same, higher-level wrapper

Env vars¶

Var	Default	Purpose
`PORT`	`8033`	HTTP listen port
`BATCH_SIZE`	`32`	Molecules per inference batch
`USE_GPU`	`auto`	Force `cpu` or `gpu`; `auto` detects at boot

Speed¶

Single molecule: ~50 ms (CPU) / ~10 ms (GPU)
Library screen (10,000 compounds): ~5 min (CPU) / ~30 s (GPU)