Discover your next lead in one connected workspace.
KemJet builds AI powering next-generation therapeutics.
KemJet is invite-only while we build it with our first users. Create an account and we will email you when you are accepted.
- 75+
- trained models
- 49+
- endpoints
- 800+
- structural alerts
Build the series, not the compound.
Sketch it, paste SMILES, drop an SDF. Then enumerate R-groups, hop the scaffold, swap in bioisosteres and grow linkers across the vector you are working on. Every candidate is a real structure, scored before anyone touches a flask.
Thirty endpoints, in the time it takes to draw.
Absorption, distribution, metabolism, clearance, toxicity and off-target risk, from checkpoints trained on scaffold splits. Each answer carries its uncertainty, whether the compound is inside the model's domain, and the fragment that moved the call.
Find the problem before it is expensive.
Eight hundred structural alerts across PAINS, Brenk, NIH and ZINC. Valence, tautomer and salt validation on import. And a counterfactual search that reports when no single substitution on the scaffold can clear your threshold.
Take it to the protein.
Dock into any PDB entry, or an ensemble of them with the spread reported as its own column. Read interaction fingerprints across a whole series and see exactly which contacts separate the actives from everything else.
Know it can be made.
Retrosynthetic search down to purchasable building blocks, every transform scored against literature precedent, with the step most likely to fail named before a single reagent is ordered.
A research assistant with your canvas open.
Palba sees what you see: every structure on the canvas and every result attached to it. It searches Europe PMC for the precedent behind a scaffold, runs the models you would have run, and carries a DOI for anything it claims from the literature. It proposes; you approve.
“No single substitution reaches 0.50 on hERG. Twenty-four candidates tried; the closest is Cl to OH at 0.68, which also costs you 0.21 of Caco-2.”