As far as I know Kb (=1 / Kd) is the equilibrium constant. Your analysis will be valid if you have the forward and backward rate constants (Kon and Koff,respectively). It seems that you are confusing Kb with Kon and Kd with Koff.

Autodock normally gives output lines like this for ligands:

USER    Estimated Free Energy of Binding    =  -12.94 kcal/mol  [=(1)+(2)+(3)-(4)]USER    Estimated Inhibition Constant, Ki   =  329.20 pM (picomolar)  [Temperature = 298.15 K]

At least up to the 4.2 version that I am using.

So, it gives ΔGbind, with negative sign. This is OK, since binding should be favorable if you have a good ligand.

The constant is a Ki, which corresponds to a Kd, NOT a Kb.

This constant is already calculated from RT ln(Ki)=ΔGbind.

Your ΔGbind has negative sign, so ln(Ki) has negative sign, Ki < 1; meaning that equilibrium is in favor of complex formation / ligand binding.

I have no idea how you can deduce kon and koff rates from the Ki; Kd, Kb, or whatever you call it. Ok, Kb=kon/koff, but this means that the equilibrium constant only gives you the ratio between kon and koff. Not their actual value.

By the way, if you do the ratios of the kon and koff values that you mention, you're unlikely to end u with a value of 40.6 µM.

So...

Paul, some very good points....

Thing is that the people at Autodock developed their scoring function with some analogy to processes happening when the ligand binds. So, you have in the end, for each ligand, terms of the scoring function such as:

USER    (1) Final Intermolecular Energy     =  -11.88 kcal/mol

USER        vdW + Hbond + desolv Energy     =   -9.12 kcal/mol

USER        Electrostatic Energy            =   -2.76 kcal/mol

USER    (2) Final Total Internal Energy     =   -0.59 kcal/mol

USER    (3) Torsional Free Energy           =   +2.39 kcal/mol

USER    (4) Unbound System's Energy  [=(2)] =   -0.59 kcal/mol

So, they have a term for the unbound system, a term for desolvation, a term for ligand torsions, and so on. And the scoring function was fitted against a number of complexes/ligands for which binding data was available. As a result of the exercise, the program provides an estimated binding constant... Below some text from their website:

But, the issue is just that the terms that are used cannot realistically account for what happens, in particular for the unbound states. Your remark about entropy is a very good one. In particular, I'd like to see how a scoring function would handle the loss of entropy resulting from conformational restriction in the receptor, upon binding of a ligand.... But (de)solvation, induced fit effects and so on are also an issue.

So, in the end, I think that the biggest problem is that people still take results from a scoring function as if it was true binding data...

In complement to what I just wrote here is some text from the Autodock website. 

"AutoDock 4.2 now has a free-energy scoring function that is based on a linear regression analysis, the AMBER force field, and an even larger set of diverse protein-ligand complexes with known inhibition constants than we used in AutoDock 3.0. The best model was cross-validated with a separate set of HIV-1 protease complexes, and confirmed that the standard error is around 2.5 kcal/mol. This is enough to discriminate between leads with milli-, micro- and nano-molar inhibition constants."

So, standard error is 2.5 kcal/mole.... around 1.7-1.8 log in the estimation of binding constants.

Another interesting bit:

"Because the scoring functions used by AutoDock 4 and AutoDock Vina are different and inexact, on any given problem, either program may provide a better result."

So, I do think that the website itself states clearly enough that the results from the scoring function should not be taken too literally and that there are limits to the interpretation one can make of those Ki / Binding energy values. It's just that people tend to forget such warnings... :-)