While many modelers know Dr. Shelley as the product manager for MacroModel, Dr. Shelley has also been acting as product manager for Desmond, a high performance molecular dynamics package developed by D. E. Shaw Research and scheduled for release as part of Schrödinger Suite 2008. Here, Dr. Shelley discusses the advances conceived and implemented by the D. E. Shaw Research team, as well as the design and implementation of the practical, straightforward Desmond interface within Maestro.
Desmond, a new parallel molecular dynamics program created by D. E. Shaw Research, provides researchers with the tools to simulate biomolecular systems with unsurpassed speed and accuracy. Schrödinger is releasing Desmond commercially and has integrated the software into the Maestro molecular modeling environment. The result is a product that makes setting up, running, and analyzing molecular dynamics simulations straightforward, allowing users to focus on the scientific problem of interest.
Desmond is a state-of-the-art molecular dynamics package, incorporating the following key innovations:
- Exploiting advanced parallel algorithms, Desmond achieves unprecedented scalability on both high-end and typical Linux clusters.
- Desmond takes advantage of SIMD (single-instruction multiple data) and other low-level parallelism strategies, providing high performance even at the single processor level.
- Desmond excels in numerical accuracy. Simulations can be run in single precision without compromising numerical rigor. Checkpoint and restore facilities are bitwise accurate, and bitwise reversible simulations are possible.
Key functionalities include:
- Explicit solvent simulations with periodic boundary conditions using cubic, orthorhombic, truncated octahedron, rhombic dodecahedron, and arbitrary triclinic simulation boxes
- Efficient parallel minimizer application for system setup and equilibration
- Ability to calculate absolute and relative solvation free energies
- Ability to calculate absolute and relative binding free energies
Key technical features include:
- Long range electrostatics rigorously treated via particle mesh based Ewald techniques
- Time-scale splitting and efficient integration using multiple time-step (RESPA) based integrators
- Accurate implementation of constraints to remove highest frequency vibrations and increase the integration time step
- Support for a variety of thermodynamic ensembles including NVE, NVT, NPT with
- Berendsen or Nose-Hoover thermostats
- Berendsen or Martyna-Tobias-Klein barostats
- Support for isotropic, semi-isotropic and anisotropic pressure coupling
- Ability to restrain the position of selected atoms
- Accurate restarts using sophisticated checkpointing facilities
- Plug-in capability to enable visualization and analysis of trajectories using VMD
D. E. Shaw Research has published a number of papers related to Desmond, including an early overview of the Desmond software1. A number of other papers describe algorithms and implementation techniques that significantly accelerate parallel MD simulations compared with current state-of-the-art codes. This includes a set of papers outlining novel parallel decomposition methods that greatly reduces the requirement for inter-processor communication2,3,4. Desmond also excels in numerical accuracy, an example of which is provided in a short note on the handling of constraints in molecular dynamics simulations5. Finally, a number of application studies using Desmond have been published. These papers include a study of the function of a transmembrane (sodium/proton antiporter) protein in which simulation results were subsequently supported by experimental site-directed mutagenesis studies6, as well as the use of microsecond-long molecular dynamics simulations to compare simulation results to those obtained by experimental NMR7.
Desmond calculations can be set up, run and reviewed from within Maestro, thereby providing an intuitive and productive environment for performing molecular dynamics simulations. New panels accessible from Maestro include the System Builder, the Desmond panel, Free Energy Perturbation Setup, Trajectory Viewer and Simulation Quality analysis. The System Builder (Figure 1) makes it easy to construct systems suitable for simulation using periodic boundary conditions starting from a solute structure; a process which involves selectively adding the molecules in the environment and assigning force field parameters (OPLS_2005). Bulk solvent (optionally containing salt) and membrane environments are supported, as are a variety of simulation cell shapes. Charge neutralization is largely automated; however, control over the placement of neutralizing ions is supported in a convenient manner. The panel is designed such that it is often sufficient just to click Start.
The Desmond panel, shown in Figure 2, is used to actually launch Desmond jobs. This panel supports starting or continuing both minimizations and simulations. The user can intuitively examine and adjust many key calculation parameters, and any coupled parameters are automatically adjusted when changes are made. The user can also elect to prepend a fully automated relaxation process to the production simulation with just a single click. As with the System Builder, the default parameters are selected such that the user need only enter the desired simulation length and click Start.
Free energies are key properties of interest in molecular systems. Free energy calculations are often tedious, difficult and time consuming to design and setup. The Free Energy Perturbation (FEP) Setup panel (Figure 3) makes setting up FEP jobs easy so that the user can focus on the real problem of interest rather than the details of the calculation. This panel supports the following solvation free energy and binding free energy calculations:
- ligand side-chain mutation (relative free energy)
- ring atom mutation (relative free energy)
- ligand annihilation (absolute free energy)
Maestro also now supports viewing simulation trajectories via the Trajectory Viewer, depicted in Figure 4. Using this comprehensive tool the user can intuitively select the frame being displayed or the speed at which the simulation is viewed. Maestro's tools permit easy manipulation of the representation and visibility of the components in the system. Trajectory smoothing, superposition to an earlier frame, replication of the simulation cell, and extraction of sets of frames are all supported features. Maestro can also create images of individual frames, or movies of entire trajectories, with only a few mouse clicks.
Figure 1. Solvating a globular protein. P38 before solvation (left). The System Builder panel in default configuration (middle). Simply clicking Start solvates and adds ions to neutralize the system, yielding the configuration depicted in the image to the right.
Figure 2. The Desmond panel. Aside from the total simulation time, the default parameter settings provided in the Desmond panel are appropriate for most simulations. Minimizations or even multistage relaxations can also be launched from this panel.
Figure 3. The FEP panel makes it possible to easily set up and launch multiple FEP calculations. Here multiple molecular fragments have been selected for addition across a bond (green) in a ligand. The conformation of one of them is included in the workspace. The user can opt to interactively adjust the conformation of each selected fragment.
Figure 4. Maestro's trajectory-viewing facility is both feature-rich and convenient.
Figure 5. The Simulation Quality Analysis panel makes it easy to review key quality measures from a simulation.
Desmond molecular dynamics simulations can be monitored from Maestro's Workspace. Desmond jobs are launched using Schrödinger's Job Control facility; providing a standardized mechanism for running both local and remote jobs, monitoring jobs, interacting with queues, halting jobs, and incorporating results into Maestro projects automatically. In addition, a plug-in for viewing Desmond simulations in VMD is provided.
Python is central to Desmond-related tasks. Nearly all Desmond jobs can be launched from Python, thus permitting the development of user-specific workflows. Analysis of Desmond simulations also makes use of Python; furthermore, straightforward and customizable scripting is possible via support from NumPy for complex mathematical operations and MatPlotLib for plotting. There also is an intuitive tool, the Simulation Quality Analysis panel (Figure 5), for assessing key markers of simulation quality.
Problem-solving often requires use of a wide range of modelling techniques. A compelling feature of this new product is the synergies that result from Desmond's integration into Schrödinger's premier molecular modelling suite for drug development. For instance, careful chemical and conformational preparation of molecules often represents a critical step in setting up a molecular dynamics simulation. The Protein Preparation Wizard, LigPrep (ligand structure) and Epik (ligand protonation state) can be used to ensure that the structures provided to Desmond are chemically correct. Starting conformations for Desmond simulations can originate from a number of sources; including, homology models produced by Prime or docked ligand poses from Glide.
Desmond's efficient and accurate simulations make it possible to thermally relax, refine, and sample conformations of proteins and protein-ligand complexes on useful time scales. Desmond's simulations may be used to select systems or conformations for further study. For example, the use of alternate protein conformations in subsequent docking calculations with Glide, or as input into SiteMap to characterize protein active sites. Another new product from Schrödinger, WaterMap, utilizes a specific type of Desmond simulations to identify and thermodynamically characterize water molecule positions within the binding site of a protein.
Desmond's accuracy, speed, scalability, and ease of use make it possible for researchers to move beyond the complexities and intricacies of setting up and running molecular dynamics simulations so that they can concentrate on the problem of interest. This should greatly expand the productive use of molecular dynamics simulations in the pharmaceutical industry.
[1] Scalable algorithms for molecular dynamics simulations on commodity clusters, Bowers, K.J., Chow, E., Xu, H., Dror, R.O., Eastwood, M.P., Gregerson, B.A., Klepeis, J.L., Kolossvary, I., Moraes, M.A., Sacerdoti, F.D., Salmon, J.K., Shan, Y., Shaw, D.E., Proceedings of the ACM/IEEE Conference on Supercomputing (SC06), Tampa, Florida, 2006, which received Best Paper Award for that meeting.
[2] A fast, scalable method for the parallel evaluation of distance-limited pair wise particle interactions, Shaw, D.E., J. Comput. Chem., 2005, 26: 1318.
[3] The midpoint method for parallelization of particle simulations, Bowers, K.J.; Dror, R.O.; Shaw, D.E., J. Chem. Phys., 2006, 124, 184109.
[4] Zonal methods for the parallel execution of range-limited N-body simulations, Bowers, K.J.; Dror, R.O.; Shaw, D.E., J. Comput. Phys., 2007, 221, 303.
[5] A common, avoidable source of error in molecular dynamics integrators, Lippert, R.A.; Bowers, K.J.; Dror, R. O.; Eastwood, M.P.; Gregersen, B.A.; Klepeis, J.L.; Kolossvary, I.; Shaw, D.E., J. Chem. Phys., 2007, 126, 046101.
[6] Mechanism of Na+/H+ antiporting, Arkin, I.T., Xu, H.; Jensen, M.Ø.; Arbely, E.; Bennett, E.R.; Bowers, K.J.; Chow, E.; Dror, R.O.; Eastwood, M.P.; Flitman-Tene, R.; Gregersen, B.A.; Klepeis, J.L.; Kolossváry, I.; Shan, Y.; Shaw, D.E., Science, 2007, 317, 799.
[7] Microsecond molecular dynamics simulation shows effect of slow loop dynamics on backbone amide order parameters of proteins, Maragakis, P., Lindorff-Larsen, K.; Eastwood, M.P.; Dror, R.O.; Klepeis, J.L.; Arkin, I.T.; Jensen, M.O.; Xu, H.; Trbovic, N.; Friesner, R.A.; Palmer III, A.G.; Shaw, D.E., J. Phys. Chem. B, 2008, [Epub].