EDKG-DL

English | 简体中文

Deep Learning framework with causality-integrated Endocrine Disruption Knowledge Graph. Predicts endocrine-disrupting effects (EDC), applicability domain (AD), and sensitive event pathways from compound SMILES using machine learning and an AOP network.

Pipeline: PaDEL 2D descriptors/fingerprints → qualitative/quantitative event-endpoint models → dual-track EDC classification (tabular model + graph neural network, GCN) → sensitive event and AOP pathway inference, producing JSON / Excel reports. Results are for research support only and must not replace experimental or regulatory conclusions.

Installation

Install from PyPI (Python ≥ 3.10):

pip install edkg-dl
# or
uv add edkg-dl

Optional extras: pip install "edkg-dl[api]" (HTTP service) and pip install "edkg-dl[plots]" (applicability-domain plots). See installation docs for details, including the automatic Temurin JRE download for PaDEL.

Model assets (~930 MB) are fetched automatically from Hugging Face on first use — no manual setup required.

Quick start

CLI:

edkg-dl-predict "CCO"                                   # JSON to stdout
edkg-dl-predict "CCO" -o runs/example --ad-plots        # reports + PCA plots
edkg-dl-predict batch molecules.txt -o runs/batch       # batch from a text file

Python:

from edkg_dl import Predictor

predictor = Predictor.from_assets("./models")
result = predictor.predict("CCO")

Documentation