scoring function

A scoring function is the judge that gives each docked pose a number, much like a referee assigning points to a gymnastics routine. The number is meant to rank how good the binding is, so that better-fitting molecules float to the top of a list.

Scoring functions come in a few flavors. Force-field-based ones add up physical interaction energies such as van der Waals contacts and electrostatics. Empirical ones sum weighted terms for features like hydrogen bonds, hydrophobic surface, and rotatable bonds, with weights fitted to known binding data. Knowledge-based ones derive statistical potentials from how atoms are observed to pack in many solved structures. Machine-learning scoring functions instead learn the mapping from features to affinity from large datasets.

The honest caveat is that no scoring function reliably reproduces true binding free energies, because that quantity depends on subtle solvent and entropy effects that fast scoring approximates poorly. Scoring functions are therefore good at separating plausible binders from junk and at rough ranking, but they routinely misrank closely matched compounds. More rigorous free-energy methods exist when accuracy matters more than speed.