Why might a good docking score fail to predict experimental binding?

Why

First distinguish pose prediction from affinity ranking. A scoring function may reward favorable static contacts while overlooking protonation states, water displacement, induced fit or a sterically inaccessible conformational state. Ligand conformer sampling, tautomer assignment and receptor preparation can change scores markedly.

A good score can also arise in a non-physiological pocket or from a distorted predicted structure. Evaluate sensitivity across receptor conformations, benchmark known active and decoy ligands, inspect interactions rather than selecting one top pose, and measure direct binding or functional modulation. For example, two compounds dock favorably to the same open enzyme conformation, but only one binds when a flexible loop closes in solution.

Use docking to generate mechanistic hypotheses and compound prioritization, not a quantitative experimental KD.