What is the most potential drug candidate in computer-aided drug design?

Docking score is just an indicator high dock score does not necessarily mean that the compound is going to active to the same extent. It gives us information about possible binding interactions but does not mean the practically the molecule may or may not bind to the same extent or have same binding as proved by docking results. Furthermore it may not help in finding the possible side effects of the drug candidate. But yes, it assists us in narrowing down the scaffold optimization and will save time which may be lost in synthesizing plenty of compounds and then checking their activities one by one.   

Prioritizing good target affinity scores over other parameters should not be expected to yield the best drug candidate. In selecting candidate compounds from a drug library, consider first narrowing your list of candidates by physico-chemical properties and expected toxicity in addition to the ADMET parameters you alluded to. Docking scores are usually more computationally expensive to obtain, so perform docking analyses after excluding unviable compounds using faster calculations. Where possible, use multiple methods to verify computed properties and affinity scores. Finally, keep in mind that binding affinity and activity are not the same thing. Depending on your target and compounds, you may encounter high-scoring binding modes that do not alter the activity of the target in the way you desire. Familiarity with the literature surrounding your target is important. If structural coordinates are available demonstrating the binding of a relevant ligand to your target, check to see whether your predicted docking modes approximate this relationship.