Beta
Beta
Computational (in silico) toxicology uses machine learning models trained on experimental assay data to predict safety liabilities from molecular structure alone. ToxScreen is a production-grade computational toxicology platform that returns calibrated probabilities — not black-box scores — and explicitly abstains when your chemistry is too novel to trust.
Get Started Free →The average drug costs $1–2 billion and takes 10–15 years to reach market. Safety failures, especially hERG cardiotoxicity and CYP inhibition, account for ~30% of preclinical attrition. A computational pre-screen before medicinal chemistry commits to synthesis can flag problem chemotypes in seconds — before weeks of in-vitro work.
• hERG channel blockade risk (cardiotoxicity)\n• CYP3A4, CYP2D6, CYP2C9 inhibition liability (DDI risk)\n• Composite Toxicity Index (CTI) — weighted traffic-light risk\n• Extended ADMET: bioavailability, BBB penetration, hepatotoxicity, and 15+ other properties\n• Applicability-domain badge for every prediction
ToxScreen is built on ADMET-AI (chemprop GNN) with three layers on top: isotonic calibration, Tanimoto applicability-domain abstention, and Wilson confidence intervals. All benchmarks are published — including where they fail (CYP novel-chemistry AUROC 0.47). Full methodology at /toxscreen/methods.
Medicinal chemists triaging hit lists. DMPK teams prioritizing compounds before in-vitro. CROs adding a computational tier. Academic groups screening natural product libraries. Cosmetic companies complying with the EU animal testing ban.