Computational Drug Discovery and Design

the protein structure is soaked in different concentrations of differ-

ent explicit solvents, allowing these solvent molecules to diffuse and

interact with hot spots on the protein surface. In this way, cosolvent

MD simulations have two major advantages over the traditional

binding site recognition methods, described above. First, cosolvent

MD simulations do not require a training set to run. Meaning that,

the method can be applied to all types of protein structures, with-

out a prior knowledge about similar protein structures or potential

binding sites [55]. The second and, perhaps, the most important

feature of cosolvent MD simulations is that it fully accommodates

protein flexibility, solvent effects and many other parameters during

the simulation (seeNote 5).

The used organic solvents can have different chemical proper-

ties and shapes, allowing the study of different possible interactions

with the protein and also helping in predicating the maximum

binding affinity for any identified binding site [1, 7, 9, 10]. During

a typical simulation the different solvent molecules are spontane-

ously distributed and concentrated around possible binding sites.

The elapsed time for solvent molecules to occupy the binding site is

directly related to its druggability [9]. In this context, the identified

binding sites are ranked by the occupation time and the increase in

the local density of the interacting organic molecules. The drugg-

ability is also assessed by the maximum binding affinity as predicted

Fig. 5The free-energy surface of protein. The local and global minimums are separated by free-energy
barriers. During the MD simulations, conformations of targeted protein are sampled. The conformations
cannot pass some large barriers and have to gather around local minimums. Because protein does not reach
global minimum, the MD simulations are insufficient sampling

Prediction of Druggable Binding Sites 95