PoseBusters
PoseBusters is a benchmark and validation suite for checking whether generated protein-ligand poses are physically and chemically plausible. It is useful when a model appears strong by aggregate score but may produce invalid structures.
Source:
Main Idea
Pose quality should not be judged only by a model score. Docking and generative workflows need explicit checks for molecular geometry, steric clashes, bond validity, and protein-ligand consistency.
PoseBusters is useful because it shifts the question from “did the model output a pose close to a reference?” to “is the generated complex physically and chemically usable enough to evaluate further?”
A useful way to think about this is a conjunction of checks:
Here is a generated pose or complex and each is a plausibility check. A pose can fail even if one aggregate score looks good.
Why It Matters
- Helps diagnose failures in protein-ligand docking.
- Separates geometric validity from scoring function performance.
- Provides a public reference point for evaluating generated poses.
- Encourages evaluation that includes both native-like binding mode and pose plausibility.
What It Checks
The exact checklist depends on the benchmark configuration, but the important categories are:
- Ligand identity and chemical consistency.
- Bond lengths, stereochemistry, and aromatic ring planarity.
- Intramolecular strain and clashes.
- Protein-ligand clashes and unrealistic overlap.
- Whether a pose is native-like only after it also passes plausibility checks.
Reading Questions
- Which generated poses fail because of ligand chemistry rather than docking location?
- Does a method report RMSD, validity, and success rate as separate quantities?
- Does the benchmark test generalization beyond training-like proteins or ligands?
- Are classical docking baselines included, and under what input assumptions?
- Is any post-processing or energy minimization applied before evaluation?
Connection to This Wiki
For this site, PoseBusters is the first evaluation anchor for Structure-Based Modeling. It should be linked whenever a docking, pose generation, or protein-ligand structure model claims good structural performance.