← research

ScreenServer

checking backend…

Screen any molecule or small library for ADMET liabilities and drug-likeness in seconds — instant in-browser descriptors plus 13 trained ML endpoints, with honest scaffold-split metrics, applicability-domain flags, and structural alerts. Free, open, CPU-only.

Who it's for: medicinal chemists prioritizing analogs, chemical biologists sanity-checking probe compounds, and academic / small-biotech drug-discovery labs that need rigorous triage on a laptop — no GPU, no per-seat license.

SMILES — one per line to screen a small libraryloading RDKit…
aspirin caffeine ibuprofen astemizole paracetamol + load all 5 as a library
The 13-endpoint ML panel
Aqueous solubility (logS) Lipophilicity (logD) Caco-2 permeability Plasma protein binding Hepatocyte clearance Acute tox (LD50) BBB penetration hERG block CYP3A4 CYP2D6 CYP2C9 Ames mutagenicity DILI hepatotoxicity
Example use cases

Hit triage

Paste a vendor catalog or hit list, rank by developability score, and drop compounds with hERG / Ames / DILI flags or PAINS substructures before ordering.

Analog prioritization

Get instant descriptors and Ro5 / QED for a proposed analog, then run the full panel to compare solubility, permeability, CYP and clearance across the series.

Probe-compound check

Confirm a chemical-biology tool compound isn't a frequent hitter and has tolerable cytochrome and cardiotoxicity risk before cellular assays.

Instant preview runs RDKit-WASM entirely in your browser — molecule render, descriptors, and Lipinski / Veber / Ghose / QED drug-likeness, with no data leaving the page. Full ML ADMET runs the real 13 trained endpoints on the backend with PAINS / Brenk / NIH structural alerts, applicability-domain checks, prediction confidence, and a composite developability score. Honest random- and scaffold-split held-out metrics are shown at the top of every report.
Open-source & CPU-only · data: MoleculeNet/DeepChem & Therapeutics Data Commons · github.com/gianyrox/screenserver · example report →
Research-grade ML triage tool. Predictions are computational estimates for prioritization, not clinical or regulatory determinations.