507
15 (How to) Profit from Molecular Dynamics-based Ensemble Docking
sampling of the receptor free energy landscape will influence results of ensemble docking. This technical advance and the use of enhanced sampling methods
(see Sect. 15.2.1.3) facilitate coverage of dynamical events such as domain rearrangements associated with slower kinetics [12].
15.2.1.2 Setting up an MD Simulation to Generate an Ensemble for Docking
Molecular dynamics simulate the evolution of an atomic system via a numerical
solution of Newton’s law of motion. The potential energy is evaluated at every time
step by a molecular mechanics force field (e.g., AMBER force field FF99SB-ILDN
[13]). Small organic molecules can be incorporated into simulations after fitting
charges from quantum mechanical calculations and atom typing via the Generalized
Amber Force Field GAFF [39].
The starting configuration of the system is generated from an X-ray or NMR
structure of the (bio)-molecule and a surrounding water shell is added. A time interval according to the fastest movements in the system (hydrogen positions)—
typically in sub-femtosecond scale—is chosen allowing for a numerical integration
over the differential equation for the particle movements. Hence, after every time
step new positions, new forces and new velocities are calculated for each atom of
the system. After a plethora of repeats a so-called trajectory of the system was successfully generated containing an in silico image of biomolecular movements in
solution.
In a perspective article from 2008, molecular dynamic simulations are pointed
out to be an “expert system” [22]. The authors highlighted that this technique is
challenging with a lot of practical caveats to be considered in system preparation
and simulation protocol. Moreover, the simulation data need to be followed and
checked for reasonable credibility of the results. As all atomic coordinates as well
as the system energy are stored, several analysis techniques can be applied after
performing molecular dynamics simulations. Standard analyses include the calculation of one-dimensional root mean square deviation (RMSD) to a reference structure yielding measures for stability of the simulation (see Fig. 15.1d). Additionally,
two-dimensional RMSD analysis reflects convergence of the simulations (compare
[33, 36, 40]).
Some of the questions when setting up the simulation are: Should a ligand-bound
or a ligand-free simulation serve as starting structure? Should the ligand be included
in the simulation? Which part of the protein should be treated as flexible? How to
handle protonation and solvation? Considering the high computational costs to run
the MD simulation an appropriate selection and preparation of the starting coordinates is essential.
Ideally, the impact of alternative starting structures will compensate during the
simulation when the conformational space is sampled exhaustively. Following the
conformational selection paradigm the ensemble of a ligand-free simulation of a receptor should cover a conformation prone to ligand binding. In agreement with this
concept, the McCammon group selects ligand-free starting structures to generate
15 (How to) Profit from Molecular Dynamics-based Ensemble Docking
sampling of the receptor free energy landscape will influence results of ensemble docking. This technical advance and the use of enhanced sampling methods
(see Sect. 15.2.1.3) facilitate coverage of dynamical events such as domain rearrangements associated with slower kinetics [12].
15.2.1.2 Setting up an MD Simulation to Generate an Ensemble for Docking
Molecular dynamics simulate the evolution of an atomic system via a numerical
solution of Newton’s law of motion. The potential energy is evaluated at every time
step by a molecular mechanics force field (e.g., AMBER force field FF99SB-ILDN
[13]). Small organic molecules can be incorporated into simulations after fitting
charges from quantum mechanical calculations and atom typing via the Generalized
Amber Force Field GAFF [39].
The starting configuration of the system is generated from an X-ray or NMR
structure of the (bio)-molecule and a surrounding water shell is added. A time interval according to the fastest movements in the system (hydrogen positions)—
typically in sub-femtosecond scale—is chosen allowing for a numerical integration
over the differential equation for the particle movements. Hence, after every time
step new positions, new forces and new velocities are calculated for each atom of
the system. After a plethora of repeats a so-called trajectory of the system was successfully generated containing an in silico image of biomolecular movements in
solution.
In a perspective article from 2008, molecular dynamic simulations are pointed
out to be an “expert system” [22]. The authors highlighted that this technique is
challenging with a lot of practical caveats to be considered in system preparation
and simulation protocol. Moreover, the simulation data need to be followed and
checked for reasonable credibility of the results. As all atomic coordinates as well
as the system energy are stored, several analysis techniques can be applied after
performing molecular dynamics simulations. Standard analyses include the calculation of one-dimensional root mean square deviation (RMSD) to a reference structure yielding measures for stability of the simulation (see Fig. 15.1d). Additionally,
two-dimensional RMSD analysis reflects convergence of the simulations (compare
[33, 36, 40]).
Some of the questions when setting up the simulation are: Should a ligand-bound
or a ligand-free simulation serve as starting structure? Should the ligand be included
in the simulation? Which part of the protein should be treated as flexible? How to
handle protonation and solvation? Considering the high computational costs to run
the MD simulation an appropriate selection and preparation of the starting coordinates is essential.
Ideally, the impact of alternative starting structures will compensate during the
simulation when the conformational space is sampled exhaustively. Following the
conformational selection paradigm the ensemble of a ligand-free simulation of a receptor should cover a conformation prone to ligand binding. In agreement with this
concept, the McCammon group selects ligand-free starting structures to generate