The 34 endpoints we predict

Never. Client structures are kept strictly separate from our training data and are not used to improve our models, under written confidentiality terms agreed before you send anything.

ADME Hero is a decision-support tool for the discovery stage, not a regulatory submission package. Filings such as an IND or CTA require GLP-compliant laboratory and in vivo studies under ICH M3(R2) and equivalent guidance, and no in silico method substitutes for them. ADME Hero's role sits earlier: it ensures that when you commission those studies, you commission them on the molecules most likely to pass.

No, and we wouldn't claim it. It decides which compounds are worth putting into them. Screening a series in silico and sending only the survivors to the bench is where the saving comes from — not from skipping the bench.

No. You send structures; we return a report written for medicinal chemists, DMPK scientists and project leads. All modelling happens on our side, and a specialist walks you through your first report.

Chemical structures (SMILES or SDF), and if you want indication-matched benchmarking, the therapeutic area you're working in. Nothing about your assays, your formulations or your programme is required.

Performance is reported per endpoint, not as one headline figure, because endpoints differ in how well they're modelled. Across our panel, agreement with the corresponding laboratory assays runs at 70% - 95%. 

The panel is built around human-relevant endpoints wherever human data exists: human intestinal absorption, oral bioavailability, plasma protein binding, human microsomal and hepatocyte clearance, and the human CYP isoforms. Where an endpoint refers to a model system — Caco-2, MDCK, PAMPA — the report says so explicitly, because that is how the reference assay is defined.

Performance is reported per endpoint rather than as one headline number, because a solubility regression and a P-gp classification are not measured the same way. Continuous endpoints are reported with correlation and mean absolute error against held-out experimental data; classifications with agreement rates. Across the panel, classification agreement runs at 70%–95%.

Predictions are strongest inside the chemical space the models were trained on. For unusual structures, confidence drops — and we flag when a molecule falls outside that space rather than reporting a number as if it were certain. Those compounds are the right candidates for early experimental confirmation.

It gives you the two predictions that decide it together: blood–brain barrier permeability and P-glycoprotein substrate status. A compound that crosses but is actively pumped back out is a common and expensive surprise in CNS programmes, and it is visible on a screen.

Yes. Skin penetration rate, solubility, lipophilicity and pKa are directly useful for cosmetic, topical and agrochemical development, and the metabolism panel matters wherever a compound will meet a liver.

They are complementary. ADME Hero predicts where a molecule goes; ToxSpotter predicts what harm it may cause. Together they give a full ADMET profile in a single submission, and most clients eventually run both.

Processing is fast; the schedule is set by scope and agreed upfront. Most engagements deliver in days rather than the months an equivalent assay campaign would take. A specialist then walks you through the report, and you keep both the visual document and the full data table.