506
S. von Grafenstein et al.
extending over several dozens of nanoseconds in the absence of a ligand [32]. Herein presented trajectories do not overlap in the sampling of alternative conformations
of the critical loop region. This gives a further hint that the prediction of a specific
ligand-bound conformation needs extensive sampling (in this case exceeding 30 ns)
or enhanced sampling methods. We decided to start sampling from three alternative
starting structures which allows covering a broader conformational space. However, this strategy is only feasible when information about alternative conformations
has already been explored experimentally. Equilibration and sampling simulation
followed an established protocol [33] using Amber12 [34] and Ambertools13 for
analysis [35]. Influenza neuraminidases were sampled in their native tetrameric
state [36].
15.2 Molecular Dynamics-based Ensemble Docking
15.2.1 Sampling the Free Energy Landscape
15.2.1.1 Principles of Molecular Dynamics Simulations
MD simulations provide a complete atomistic view of (bio-)molecular motions on
the femtosecond to microsecond scale. The free energy landscape determines which
states will contribute to an ensemble of structures at a given temperature. Also kinetic aspects, e.g., transition frequencies between different states are determined
by the energy barriers within the free energy landscape of the system [12]. Molecular dynamics simulations explore the landscape given an energy distribution
determined by the system’s temperature.
First computer simulations of a protein system were described in 1977 [37] with
a trajectory length of 9 ps in vacuo. Within this time scale side chain movements
could be observed, whereas the backbone geometry remained virtually unaffected.
Since then continuous increase in computing power allowed extension of sampling
time. Time scale of most publications we will refer to is in the nanosecond range.
Loop movements including fast domain motions can be observed within this time
scale [12]. There is evidence that increasing sampling time positively affects the results of subsequent docking to the structural ensemble: the fraction of ligand poses
similar to the native pose increased for the test cases thrombin and acetylcholinesterase, when comparing a 50 ps simulation with a 30 ns simulation [38]. Also, single
observations of transitions between ligand-bound and free state were observed in
MD simulations of this range, e.g., for influenza neuraminidase in a 100 ns simulation [32]. However, statistical sampling of transition rates between these states is
mostly not reached and we must expect that the simulations discussed within this
overview do not cover larger loop movements.
Current simulation techniques are capable of routinely capturing molecular motions in the microsecond time scale covering water molecules as explicit solvent
by usage of GPUs [15]. It will show up in the near future how the more thorough
S. von Grafenstein et al.
extending over several dozens of nanoseconds in the absence of a ligand [32]. Herein presented trajectories do not overlap in the sampling of alternative conformations
of the critical loop region. This gives a further hint that the prediction of a specific
ligand-bound conformation needs extensive sampling (in this case exceeding 30 ns)
or enhanced sampling methods. We decided to start sampling from three alternative
starting structures which allows covering a broader conformational space. However, this strategy is only feasible when information about alternative conformations
has already been explored experimentally. Equilibration and sampling simulation
followed an established protocol [33] using Amber12 [34] and Ambertools13 for
analysis [35]. Influenza neuraminidases were sampled in their native tetrameric
state [36].
15.2 Molecular Dynamics-based Ensemble Docking
15.2.1 Sampling the Free Energy Landscape
15.2.1.1 Principles of Molecular Dynamics Simulations
MD simulations provide a complete atomistic view of (bio-)molecular motions on
the femtosecond to microsecond scale. The free energy landscape determines which
states will contribute to an ensemble of structures at a given temperature. Also kinetic aspects, e.g., transition frequencies between different states are determined
by the energy barriers within the free energy landscape of the system [12]. Molecular dynamics simulations explore the landscape given an energy distribution
determined by the system’s temperature.
First computer simulations of a protein system were described in 1977 [37] with
a trajectory length of 9 ps in vacuo. Within this time scale side chain movements
could be observed, whereas the backbone geometry remained virtually unaffected.
Since then continuous increase in computing power allowed extension of sampling
time. Time scale of most publications we will refer to is in the nanosecond range.
Loop movements including fast domain motions can be observed within this time
scale [12]. There is evidence that increasing sampling time positively affects the results of subsequent docking to the structural ensemble: the fraction of ligand poses
similar to the native pose increased for the test cases thrombin and acetylcholinesterase, when comparing a 50 ps simulation with a 30 ns simulation [38]. Also, single
observations of transitions between ligand-bound and free state were observed in
MD simulations of this range, e.g., for influenza neuraminidase in a 100 ns simulation [32]. However, statistical sampling of transition rates between these states is
mostly not reached and we must expect that the simulations discussed within this
overview do not cover larger loop movements.
Current simulation techniques are capable of routinely capturing molecular motions in the microsecond time scale covering water molecules as explicit solvent
by usage of GPUs [15]. It will show up in the near future how the more thorough