What is a docking score?

After docking generates poses, a scoring function assigns each one a number. In AutoDock Vina and many similar tools the score is expressed in kcal/mol, and more negative means better: a pose scoring −10 kcal/mol is predicted to bind more tightly than one scoring −7.

What the number means — and what it does not

The units suggest a binding free energy, but a docking score is an approximation built for speed. It typically ignores or simplifies solvation, protein flexibility, entropy and electronic effects. Scores are therefore:

• useful for separating likely binders from unlikely ones in a large library;
• unreliable for ranking close analogues whose true affinities differ by less than about ten-fold;
• not comparable between different programs, scoring functions, or protein targets.

Deep-learning docking methods such as DiffDock report a confidence instead — how likely the pose is to be correct — which is a different quantity again.

The size bias

Bigger molecules make more contacts, so they tend to score better almost regardless of fit. To compare compounds of different sizes, chemists normalise by size using ligand efficiency (LE):

LE ≈ −score / number of heavy atoms

Experimental LE is calculated the same way from measured binding free energy; values around 0.3 kcal/mol per heavy atom or higher are generally considered good starting points. Size-normalised metrics are especially important when comparing macrocycles with small molecules.

Reading results sensibly

1. Look at the score distribution of the whole screen, not only the best value.
2. Check the top poses for clashes and unsatisfied polar groups.
3. Prefer compounds whose top few poses agree with each other.
4. Where possible, include known actives and inactives as controls.