How to tell if our ligand-protein docking is good from AutoDock Vina's result

Question

I have perform a ligand-protein docking using Autodock Vina.
The result of the docking looks like this:
WARNING: The search space volume > 27000 Angstrom^3 (See FAQ)
Detected 8 CPUs
Reading input ... done.
Setting up the scoring function ... done.
Analyzing the binding site ... done.
Using random seed: -1553787135
Performing search ... done.
Refining results ... done.

mode |   affinity | dist from best mode
     | (kcal/mol) | rmsd l.b.| rmsd u.b.
-----+------------+----------+----------
   1         -5.9      0.000      0.000.  Pose 1
   2         -5.7     22.945     25.492.  Pose 2
   3         -5.5      1.426      2.046.  Pose 3
   4         -5.5     23.669     25.616
   5         -5.4     25.783     29.152.  .....
   6         -5.3     21.146     23.357
   7         -5.2     20.323     22.545
   8         -5.2     23.864     26.064
   9         -5.1     23.422     26.585.  Pose 9

As far as I understand from these statistics Mode 1(Pose 1) is the best.
However when I actually visualize them in Pymol, Pose 1 has no hydrogen bonding at all
but Pose 2 has.
My question is how can we judge if which of those two Pose is the best to use?
Note in figure below Pose 2 has dashed line (Hydrogen bond).

Matteo Ferla · Answer

It is best to contextualise the numbers. -1 kcal/mol is about the potential energy gained from a hydrogen bond —technically described in the r^6 part of the Lenard–Jones term, it is also the average collision energy of a water molecule at 37°C as that is RT ($frac{k_bcdot T}{N_A}$, wiki)under a Maxwell–Boltzmann distribution. A salt bridge –2 kcal/mol (Columbic force term).
So your scores are not very low, hence why you are counting two hydrogen bonds. Although you can also see a lovely sulfur-pi interaction which is –1 to –2 kcal/mol, so those bonds alone are probably making a –4 kcal/mol contribution, so I am guessing that some terms may be horrendous, such as repulsion forces etc.
A nice metric is doing a conversion to ligand efficiency, which weeds out affinities driven by size of the molecule... in this case most of the molecule is doing nothing.
Also, most programs have an accuracy of 1 kcal/mol or higher.
So one cannot say what ∆∆G is the best with the data at hand due to noise, but one can say that the ∆∆G is not low enough... Sorry.

How to tell if our ligand-protein docking is good from AutoDock Vina's result

One Answer

Add your own answers!

Ask a Question