Toxicity QSAR & Early Safety Screening

We maintain 30+ toxicity-focused QSAR models that deliver rapid, decision-grade predictions across key preclinical safety endpoints. Each model is trained on manually curated, literature-anchored datasets integrating molecular interaction networks, pathway context, gene–disease associations, xenobiotic metabolism (e.g., CYP), and historical in vitro/in vivo toxicity outcomes. Models are rigorously validated (nested cross-validation + external hold-outs), probability-calibrated, and deployed with applicability-domain and uncertainty estimates—enabling risk-aware triage rather than binary pass/fail screens. Both classification (risk flags) and regression (dose/severity surrogates such as MRTD) formulations are supported, with interpretable feature attributions to surface mechanistic alerts.

Covered endpoints include: cardiotoxicity (e.g., hERG/TdP risk), neurotoxicity, nephrotoxicity, hepatotoxicity (incl. liver weight gain), cytotoxicity, Ames mutagenicity, genotoxicity, MRTD, and skin sensitization, among others. These models plug directly into our discovery workflow: ultra-large libraries are safety-filtered up front, then advanced to structure-based evaluation (e-pharmacophore, DRGSCROLL flexible docking), MD/free-energy refinement, and medicinal chemistry optimization—reducing downstream attrition while preserving viable chemical space.