Show HN: ARR-Medic-CYP3A4 – Open-Source Drug Interaction Toolkit (github.com)

ARR‑MEDIC‑CYP3A4 is an open‑source toolkit for predicting CYP3A4 inhibition aimed at research, education, and prototype development (explicitly not for clinical use). CYP3A4 metabolizes >50% of drugs, so predicting inhibition helps flag drug–drug interactions, informs polypharmacy safety, and accelerates early‑stage screening. The project emphasizes transparency and reproducibility: a visible codebase, documented pipelines, and community contributions make it a teaching and benchmarking platform rather than a regulatory product. Technically, v1 uses rule‑based molecular descriptors (MW, LogP, TPSA, etc.) yielding ≈70% accuracy (sensitivity ~75%, specificity ~65%)—positioned as an “educational sweet spot.” The repo includes a FastAPI backend, React frontend, Docker/Conda install paths, sample data, and REST endpoints (POST /predict, /predict/batch), with latency <2s and throughput 100–500 predictions/min. The stack supports RDKit (200+ descriptors), classical ML (RandomForest, XGBoost → ~80–85%), and roadmaped GNN/Transformer models (~85–90%), with an offered commercial “Pro” path for 90%+ clinical validation. Key implications: an accessible, extensible baseline for method comparison, feature‑importance and interpretability experiments (SHAP/attention), and a migration path toward production-grade systems while keeping safe, non‑clinical experimentation and community-driven validation at its core.