Pose prediction plateaus after about five crystal structures per scaffold
- Using 403 SARS-CoV-2 main protease structures from the open COVID Moonshot campaign, the authors asked how many new crystal structures are worth collecting.
- Similar reference ligands give the best pose prediction, and success plateaus at roughly five structures per Bemis-Murcko scaffold, exceeding 95% for the lead series.
💡 Points
This gives structural biology a concrete budget line: after about five structures per scaffold, the next soak buys little for pose prediction. Caveat: Mpro is a rigid, well-behaved pocket, so I'd not carry that number over to flexible kinases or induced-fit targets.
How Many Crystal Structures Do You Need to Trust Your Docking Results? ↗