Desolvation energy is an important factor in the protein-ligand binding affinity prediction.However,on most of current scoring functions,it is often roughly estimated by some simple descriptors only from the ligand,with the considerations of the speed and robustness.In theory,the energy changes of the water molecules displaced by the ligand are more appropriate for reflecting the desolvation effect during the binding process.Starting from this idea,we developed a knowledge-based method for predicting the locations and energies of the potential water molecules within the protein binding pocket.Upon them,the desolvation energy could be calculated via a "displaced-solvent" functional.It was integrated with other energy terms from X-Score by a genetic algorithm.The new scoring function with desolvation energy demonstrated improved performance in binding affinity prediction.In addition,it indicated that such a strategy could be applied in other general scoring functions.