Discrete Molecular Dynamics Simulation of Biomolecules - Computational Modeling of Biological Systems - page 66

6.5 A, where the interaction potential attenuates in all atom pairs; (2) we assign a
potential step between the distances corresponding to the energy minimum (force is
zero) and the interaction action range (force approaching zero), where the force
is maximum; (3) we choose the hard sphere distance with VDW-EEF1 energy
equal to the minimum energy plus 2k B T 1:2 kcal=mol, since thermodynamically
the probability to find two atoms within this distance is very low. We choose the
next repulsion step with VDW-EEF1 energy equals to the minimum energy plus
k B T 0:6 kcal=mol, and the third repulsive step before the energy minimum with the
repulsive force

20 pN, a relative strong force in biology. The energy at each step
of the potential is computed as the average of the continuous VDW-EEF1 function,
except for the region corresponding to the energetic minimum.
We model the hydrogen bonding interaction using the reaction algorithm, which
has been adapted to the all-atom representation (Fig. 6 d). All possible interactions
between backbone-backbone, backbone-side chain, and side chain-side chain
atoms are included. Long-range electrostatic interactions were not included in the
previous work [ 21 ]. Recently, we have included the electrostatic interaction between
formal charges using the Debye-Huckel approximation, which results in better
prediction of protein-peptide and protein-ligand interactions (unpublished work).
Other efforts in methods development of all-atom DMD model include those
by Borreguero et al. [ 29 ], Emperador et al. [ 31 ], and Luo et al. [ 30 ]. However,
these models are either nontransferable with structure-based interaction models
[ 30 ] and constraints for specific secondary structure [ 31 ], or not systematically
benchmarked [ 29 ].

3.3 Extension of the Force Field for Small Molecules

Recently, we have extended the Medusa force field in order to model small molecule
ligands [ 42 ] by introducing new atom types and parameterizing the pairwise VDW
and EEF1 interactions. We performed a benchmark of the new force field by
predicting the binding affinities of a large set of protein-ligand complexes. The
correlation coefficient between the computational and experimental affinities is
approximately 0.6, which is comparable to other existing computational approaches.
Additionally, we developed a flexible ligand docking method using the new force
field for both ligand and pose selection [ 43 ]. The results of the docking benchmark
are comparable to or better than those of other flexible docking programs on the
market [ 43 ]. Therefore, the extended Medusa force field is useful in modeling small
molecules.
We discretized the small molecule Medusa force field extension in order to model
small molecules in DMD simulations. Using a similar discretization protocol to that
described above for VDW-EEF1, we can readily obtain the nonbonded interactions
for small molecules. Since there are an insufficient number of high-resolution
small molecule structures to determine the parameters for the bonded terms, we