Fold, find pockets, dock, triage

Best fit: you have a target (a sequence or a UniProt ID) and want to know its shape, where a molecule could bind it, and whether a candidate compound looks tractable — in one pass.

The prompt

I want the 3D shape of KRAS G12C (UniProt P01116) and every druggable pocket
on it — focus on the Switch II binding site and check whether sotorasib's
covalent binding mode is recapitulated. Then assess drug-likeness on the docked
compound and tell me which non-Switch-II pockets score above 0.5 druggability
for a non-covalent allosteric back-up.

What it does

The orchestrator runs a structure-first chain, passing each result into the next:

1. ESMFold predicts the 3D conformation from the sequence.
2. P2Rank scores druggable pockets on that structure.
3. DiffDock docks the named ligand into the pocket of interest and returns a pose with a confidence score.
4. ADMET-AI reads drug-likeness on the docked compound.

Watch for

• Low-confidence regions. pLDDT below 70 marks disordered loops or uncertain geometry — treat pockets that sit in those regions with caution.
• Pose plausibility. A high DiffDock confidence is not proof; check whether the pose recapitulates a known binding mode, as the prompt asks.
• Covalent binders. Docking scores non-covalent poses; a covalent warhead's true binding mode may not be captured by the score alone.

Read next

• Which tool? — pick a different structure or docking tool.
• Hit to lead — carry a docked hit forward to synthesis.
• Virtual screening — the guided version: library docking step by step.