Ideally, you would want to relate the calculated ddG of complex formation to the experimental ddG of binding. Predicted ddG is calculated from the difference of the dG predicted for the complex minus the sum of the dG of the isolated components.
The Gibbs free energy change (ddG, read: delta delta G) of a reaction is related to the equilibrium constant of that reaction by the formula:
delta G0=-RT ln Keq
Therefore, the more negative the dG, the more the equilibrium of the reaction is on the side of the product.
Experimentalists are used to the notion that a strong negative value is indicative of a high binding affinity.
Docking is the simpler version of the interaction of ligand and protein if the score is -ve then it states that the ligand has less sideeffect. so wwe go for -ve and minimum value.
In docking programs like GOLD, the binding affinities are evaluated by scoring functions, whose value is positive. The higher (and positive) the scoring function, the higher the potential affinity.
In molecular mechanics/molecular dynamics you subtract the energy of the complex from their components. If the result is negative, it indicates that the complex is more stable than their components by themselves.