Understanding Binding Energetics in Molecular Docking
A docking score is a balance sheet, not a binding free energy. Polar contacts have to pay back the water they displaced. A flexible ligand pays an entropy cost. A larger ligand often scores better because it makes more contacts. Read the score only after you know which of those terms the function even has.
Desolvation
Before the ligand binds, water stabilizes its polar groups and the protein surface. Binding strips some of that water. If the protein does not replace those interactions, the cost stands.
A charged amine is the example. In water it is well solvated. Buried in a hydrophobic pocket with no partner, the lost hydration shell is unfavorable even if the shape fit looks clean.
Entropy
Binding removes motion. The ligand loses translation and rotation. A flexible ligand loses conformations. Side chains at the site often order. That entropy opposes binding.
Entropy can also help. Ordered waters released into bulk solvent are more disordered than they were in the pocket. A rigid ligand pays a smaller conformational penalty than a flexible one, which is why rigidity can improve binding efficiency.
Water that was already in the pocket
Two cases:
- A high-energy, poorly stabilized water is a good thing to displace. It gains freedom in bulk solvent.
- A water with strong hydrogen bonds is a bad thing to displace unless the ligand restores those bonds.
Hydration analysis is how this note suggests finding the “unhappy” waters worth targeting.
Electrostatics, and the solvent around them
Salt bridges, hydrogen bonds, and dipoles are the favorable electrostatic terms. A carboxylate next to a lysine or an arginine is the example.
The same charge is stabilized by water. Bury it without a partner and the desolvation cost can exceed the interaction you gained. The electrostatic term and the desolvation term are one argument.
Larger ligands score better
More atoms make more contacts, so many scoring functions drift toward larger molecules. Ligand efficiency (LE) divides the binding energy by the number of heavy atoms. Fit quality (FQ) is the size-adjusted efficiency named alongside it. A smaller ligand with efficient contacts can be the better lead.
What each scorer actually computes
AutoDock4 is a physics-based force field. Partial charges are explicit, commonly Gasteiger-PEOE. Electrostatics are a screened Coulomb term. Hydrogen bonding and desolvation are separate terms, calibrated on experimental binding data.
Vina is an empirical function trained on experimental complexes. It does not compute partial charges and it has no classical Coulomb term. Steric fit, hydrophobic contact, and hydrogen bonds are empirical potentials. It is faster, and the electrostatics are less visible.
Smina is a Vina fork with refined empirical potentials, a torsional penalty, and a stronger hydrophobic term. It is still empirical and still does not compute partial charges.
GNINA adds a convolutional network trained on protein–ligand complexes. The network reads the 3D pattern and does not model electrostatics as a physics term. Pose prediction is the strength named here. Detailed charge effects are the gap.
| Function | Electrostatics | Partial charges | Speed | Strength named here |
|---|---|---|---|---|
| AutoDock4 | Physics-based | Explicit | Slower | Charged systems |
| Vina / Smina | Empirical, implicit | None | Fast | Large screens |
| GNINA | Neural network | None | Fast | Pose prediction |
How to read the number
The score mixes van der Waals, electrostatics, hydrogen bonds, desolvation, and some entropy-like penalties, and several of those pieces are empirical. Use it to rank compounds in one dataset. Do not read it as an absolute affinity.
Binding is the case where the contacts, the hydrophobic fit, and the release of a bad water outweigh desolvation, the frozen ligand, and the size bias in the function you chose.
