from these interactions [35]. The maximum binding affinity can be
obtained by calculating the associated free energy,ΔGi. It counts
the number of times the different solvents are attracted to a given
hot spot and compares with expected value.ΔGi¼kBTInðÞðNi=No 2 ÞIn Eq. (2),kBis the Boltzmann constant,Tis the temperature,
Nirepresents the observed solvent population, whileNorepresents
the expected value.
Different studies using the cosolvent MD simulation technique
helped in identifying new binding sites and identified the limita-
tions of the method. For example, in 2009, Seco et al. performed
MD simulations on five different proteins, dissolved in explicit
binary solvents, 20% isopropanol–water (volume/volume). They
were able to identify the binding sites on the target proteins and
evaluate their druggability [9]. However, they also pointed out to
the effect of the simulation time as well as the low diffusion rate of
the solvents on the binding site prediction. In particular, they
concluded that a low diffusion rate can limit solvent exchange and
can also prevent the solvent probes from diffusing into cavities and
gaps within the protein structure [9, 56].2.7 Site
Identification by
Ligand Competitive
Saturation
Guvench et al. developed a more refined MD-based method,
namely the site identification by ligand competitive saturation
(SILCS). The method uses explicit ternary solvents, comprising
benzene, isopropanol, and water. The objective of combining
these different solvent molecules was to provide a precise division
of affinity properties among hydrophobic groups, aromatic groups,
hydrogen bond donors, and hydrogen bond acceptors
[57, 58]. They validated the SILCS method using a full flexible
simulation for Bcl-6. While including different solvent molecules
improved the prediction of interaction energies, it did not over-
come the diffusions’ limitation [36]. Raman et al. brought up an
improvement to the SILCS method in 2013, where more diverse
solvent molecules were used [59].
A related approach, with some distinct technical details, is
called MixMD and involved an explicitly binary solvent, 50% acet-
onitrile–water (weight/weight). In enhanced method, Lexa et al.
added isopropanol too into the MixMD approach and tested on five
proteins. They indicated multiple short MD simulations might be
more efficient in sampling binding sites than few but long MD
simulations [58]. These two methods have been successfully
applied to ligand design and reproducing crystallographic binding
sites of small organic molecules [56].
A similar method, named MDmix, employed two binary sol-
vent systems, 20% acetonitrile–water and 20% isopropanol–water
(volume/volume). Two proteins were tested, namely, the heat96 Tianhua Feng and Khaled Barakat