From a scaffold to a calibrated clearance estimate.
The molecule-side funnel, end to end. Profile a series for ADMET and compliance, optimize the leads, dock against the target with strain correction, and gate on a calibrated Phase I clearance estimate — without leaving the assistant.
Toxicity and drug-like properties drive roughly 40 to 45 percent of clinical failures. We surface those flags in silico, before a dollar of lab spend.
What it looks like in practice.
The accuracy, stated plainly.
Phase I AUROC
SAMPL7 RMSE
models
SOTA endpoints
Pre-computed ADMET on every compound
Absorption, distribution, metabolism, excretion, and toxicity flags are already computed across 122M molecules — and predicted on demand for novel structures. No model queues, no compute allocation.
Docking with strain correction
AutoDock-GPU poses, corrected for internal ligand strain so the binding estimate survives a medicinal chemist’s scrutiny. Reference-ligand co-docking keeps the pose honest.
Lead optimization that respects the series
Scaffold-aware, property-directed optimization. Tanimoto thresholds tuned to the 0.80–0.85 sweet spot so generated analogs stay synthesizable and on-series.
A clinical-outcome gate, calibrated
NovoExpert-3 estimates Phase I clearance from physicochemical, ADMET, and target/indication features. Strong on cardiovascular and GI; we state where it is not yet validated.
Run the molecule-side funnel on an engine you can audit.
NovoMCP is open to a small group of PIs, postdocs, and research engineers. Tell us what you are working on.