A tale of two states: AlphaFold binding pockets on the apo-holo continuum

AlphaFold pocket states across 237,922 pockets from 12,255 proteins are mostly holo-like (62%), with holo-likeness tied to how experimental apo and holo structures align.

• Why it matters: Ligand binding is often inferred by comparing apo versus holo structures, but it is unclear whether AlphaFold-predicted pockets resemble apo-like or holo-like states. This limits reliable binding-site detection, docking, and interpretation of predicted models.
• What they did: The study maps AlphaFold pockets onto an apo–holo axis defined by every experimental structure of each site, then tests three binding-site predictors and cognate-ligand docking against experimental state separation. It also evaluates whether AlphaFold confidence scores indicate pocket state.
• The result: Holo-like pockets are predicted 13–21 percentage points more often by three predictors and enable docking 19 points more often, with gaps widening as apo and holo states separate and disappearing where they overlap. AlphaFold confidence reflects pocket variability rather than apo-versus-holo state, and the authors release per-pocket classifications for benchmarking and docking readiness.

AI summary of the abstract · may contain errors · not medical advice · read the paper.