be used to afford protein flexibility during the ligand binding event
but it is only limited to a small number of residues usually in the
proximity of the binding pocket. Thus, it cannot take into account
important conformational changes occurring during drug binding
(induced-fit effect) [22–25]. Furthermore, docking scoring func-
tions, which allow discriminating and ranking the compounds
according to their predicted binding affinity, are limited in accuracy
and can be often misleading [26–30]. MD is a powerful tool which
helps to overcome these critical limitations, and is generally imple-
mented before and after docking.
MD is a computational simulation of a complex biological
system which describes motions, interactions, and dynamics at the
atomic level by choosing a “force field” describing all the inter-
atomic interactions and by integrating the Newtonian equations
which give position and speed of atoms over time [31–34]. As
protein and protein–ligand motions would involve highly complex
and computationally expensive quantum mechanics (QM) calcula-
tions, MD aims at approximating these QM terms and movements
governed by probability functions through Newtonian physics and
by implementing force-fields that are parameterized to fit physico-
chemical knowledge obtained from experimental data. Before
docking, MD allows a conformational sampling and clustering of
a protein or enzyme to account for protein dynamics and the
conformational selection by a ligand [35, 36]. This allows facing
the limitations of a “static” experimental structure by docking
molecules to a representative set of protein’s conformations. In
some other cases where it is not obvious to find a proper binding
site, it allows the detection of cryptic or allosteric cavities that were
not present in the initial experimentally determined structure
[37]. MD is also used as a second-step filtering process to further
validate a protein–ligand complex obtained from docking by deter-
mining the stability of the complex from a trajectory, identifying
persistent interatomic interactions and by estimating the binding
free energy (Fig.2). This more accurate protein-complex evalua-
tion and validation allows better ranking of the hit molecules and
further reduces the time and cost during experimental testing as
well as the number of false positives and negatives [38]. Finally, by
including the induced-fit and solvent effects, we can work with a
more realistic model, which may apply even to a membrane-like
environment in the case of membrane proteins [39]. These MD
procedures will be described in detail in Subheading3.
Docking and MD approaches are highly complementary
computational methods for drug screening. But in a way similar
to docking, MD also faces its own limitations. The current issues
concern the high computational cost and approximations in the
force-fields used, even if considerable improvements have been
made in computer power and algorithmic efficiency [40–44]. Sub-
heading3 will present protocols to prepare a protein and MD files,
Molecular Dynamics in Virtual Screening 147