Your pocket, not a generic one
Molecules are designed and scored against the exact binding site you define, so candidates are shaped by your target rather than picked off a shelf.
Generative follow-up to virtual screening
Your library screen is done. Now search the chemical space it never covered.
Point ChemLlama at the same receptor and pocket you already screened. It designs new molecules for that site, docks them, and returns ranked candidates with drug-likeness and synthesizability — no models to install, no GPUs to configure, no docking pipeline to maintain.
- Reuse your existing PDB structure and binding site — no new setup to learn.
- Get candidates outside your screened library, not the same chemotypes again.
- Every molecule comes with a docking score, QED, and synthetic accessibility.
Start with your target
Enter a PDB code or upload your own structure file.
or
Look around and set up a pocket without an account — you only sign in to run.
- Any PDB target
- your structure, your pocket
- New chemical matter
- designed, not enumerated
- 3 scores per molecule
- docking · QED · synthesizability
- 0 infrastructure
- no GPUs, no local pipeline
From your screened target to new candidates
The same receptor and pocket you already worked with, four steps, and a shortlist of predicted docking candidates at the end.
- 01
Load your target
Enter the PDB code you screened against, or upload your own .pdb / .pdbqt structure.
- 02
Set the binding site
Place the docking box on the pocket you care about — guided by a co-crystal ligand or your own coordinates — and confirm it in 3D.
- 03
Let it design and dock
New molecules are proposed for your pocket and docked with QuickVina, then scored for drug-likeness and ease of synthesis.
- 04
Review the candidates
Get a ranked shortlist of predicted docking candidates to judge with your own medicinal-chemistry expertise.
What you get that a library screen cannot give you
Candidates designed for your pocket, judged on binding, drug-likeness, and synthesizability together — with a setup you can inspect before anything runs.
Beyond your screened library
A fixed library can only return what someone already made. Generative search proposes chemical matter your screen had no way to reach.
Developable, not just tight-binding
Candidates are scored for drug-likeness (QED) and synthetic accessibility alongside docking, so a promising score is not attached to an unmakeable molecule.
No infrastructure to run
No model weights, no GPU allocation, no docking pipeline to keep alive. Load a structure in the browser and start a run.
Setup you can inspect
See the structure, pocket, and docking box in 3D and confirm them before the run — the setup is yours to check, not a black box.
Grounded in open research
Built on published, openly available work from YerevaNN, so the methods behind your candidates can be read and cited.
The science behind the candidates
ChemLlama builds on open research from YerevaNN on generative molecular design and protein-aware optimization. Candidate molecules are proposed by generative models, docked against your receptor with QuickVina, and scored for drug-likeness and synthetic accessibility. Everything returned is a predicted docking candidate, not an experimental hit — the platform is there to widen the set of ideas worth your time, not to replace your judgement or your assay.
- Generative designproposes molecules for your pocket
- QuickVina dockingpredicted binding pose and score
- QEDdrug-likeness of each candidate
- Synthetic accessibilityhow hard it would be to make