Accessing ANTON 2 for COVID-19 Research
The Anton 2 special-purpose molecular dynamics supercomputer hosted at Pittsburgh Supercomputing Center (PSC) is provided without cost by D.E. Shaw Research for non-commercial use by the U.S. research community. PSC supports the community’s use of this resource with operational support from National Institutes of Health award R01GM116961.
How Anton 2 can help COVID-19 research
Anton was developed by D. E. Shaw Research to execute molecular dynamics simulations of biomolecules (e.g., proteins, nucleic acids, and lipids), and their interactions with natural signaling molecules and drug candidates, orders of magnitude faster than was previously possible. Anton’s speed provides a unique capability for fast turnaround of ultra-long-timescale molecular modeling, enabling observation of important biological events.
How to Apply
Applications for time on Anton 2 [1] are being accepted through the COVID-19 HPC Consortium for urgent work with the potential to impact the national COVID-19 response. Due to the special nature of Anton 2, critical information identified below must be addressed in Anton 2 proposals submitted to the COVID-19 HPC Consortium, in addition to the requirements specified on the COVID-19 HPC Consortium site. In particular, investigators must 1) provide relevant simulation details demonstrating that their proposed simulations satisfy the Simulation Requirements outlined below and 2) clearly explain why achieving their objectives requires access to Anton 2, and could not be efficiently achieved on other high performance computing platforms. Generally, this justification will include the need for longer MD trajectories than are feasible on conventional systems.
Consideration will be given only to applicants whose projects satisfy the Simulation Requirements outlined below. Investigators who have questions regarding the suitability of their proposed simulations are encouraged to contact grants@psc.edu to discuss their planned project before submitting the proposal.
Resources should be requested in terms of machine days on Anton 2 at PSC. Benchmarks to facilitate the estimation of simulation resources are given in the Estimating Simulation Resources below. Investigators must clearly state and provide justification based on the provided benchmarks for the number of machine days requested. Requests over 10 machine days will require exceptionally strong justification.
Anton does not run Desmond, AMBER, NAMD, GROMACS, or any other MD simulation software package, although it uses a Desmond structure (DMS) file as an initial input and its trajectory output is compatible with Desmond’s (DTR files). Each simulation must be built, using tools available at PSC, specifically for Anton. Biomolecular systems should be well equilibrated prior to running on Anton. PSC will provide instructions and staff assistance to help convert files to Anton format for equilibrated systems generated with the following programs: Desmond, Amber, CHARMM, Gromacs, or NAMD. However, for the easiest conversion, we strongly suggest using Desmond to equilibrate if possible, especially if your system has custom molecules.
Simulation Requirements
- Proposed simulations must be standard MD runs in the constant NVE, constant NVT (isothermal), or constant NPT (isothermal, isobaric) ensembles. Constant NVT and NPT simulations must use the Multigrator framework described in [2] with either Nose-Hoover or Langevin thermostats and isotropic or semi- isotropic MTK barostats. Simulation conditions may include the specification of a uniform constant applied electric field. Position restraints, on a per atom basis, are allowed. Enhanced sampling is also available in three basic forms: (i) simulated tempering with the Nose-Hoover thermostat, (ii) application of restraints between the centers of mass of groups of atoms, and (iii) application of conformational restraints, each based on the calculation of RMSD (root mean squared deviation) with respect to atomic positions of a given reference structure. For restraints in both (ii) and (iii), equilibria and spring constants can be varied during a simulation according to a schedule. Applicants with systems that have dozens of restraints and/or restraints involving thousands of atoms should contact grants@psc.edu before submitting a proposal.
- The simulation cell must have only right angles (i.e., it must be a cubic or orthorhombic box), and must be a minimum of 45 Angstroms on each side. Applicants with systems shaped such that one dimension of the simulation cell is much larger than the others should contact grants@psc.edu before submitting a proposal.
- Proposed simulations must use recent variants of the following standard, nonpolarizable biomolecular force fields: CHARMM (e.g., CHARMM22, CHARMM27 – including CMAP corrections, and CHARMM36), AMBER (e.g., AMBER99, AMBER99SB, AMBER03), or OPLS (e.g., freely available OPLS-AA/L; proprietary OPLS versions are not supported). Modified versions of the CHARMM and AMBER force fields, based on published research by DESRES, are also acceptable (and available through the simulation setup tools). Water should be modeled with the SPC, TIP3P, or TIP4P models, or their variants.
- Chemical systems proposed for simulation must contain between 25,000 and 700,000 atoms (including solvent atoms), though systems between 50,000 and 600,000 atoms are recommended for maximum efficiency. Chemical systems proposed for simulation must consist of some combination of protein, DNA, RNA, lipids, water, and standard ions. Investigators who wish to use custom parameters or molecules that are not included in the standard distribution of the supported force fields (see 3 above) should contact grants@psc.edu to discuss the suitability of their simulations before submitting their proposal.
- Anton is capable of producing long, continuous MD trajectories. To maximize the benefit of this resource to the research community, researchers should not propose to run simulation trajectories that would finish in less than one hour on Anton (e.g., less than ~2,500 ns for a 25,000 atom system, and less than ~150 ns for a 700,000 atom system; see Estimating Simulation Resources).
- State whether the proposed system has already been built and equilibrated. If the system has not yet been run in production, please provide evidence that the proposed system can be successfully built and simulated.
- Provide strong scientific arguments as to why the length and number of proposed simulation runs will be both sufficient and necessary to achieve the stated scientific objectives.
Estimating simulation resources
Please refer to the table below, with benchmarks for a number of systems of various sizes, to estimate the amount of machine time required for the project. The actual achievable simulation times may vary even for different molecular systems of similar size. Each Anton 2 simulation runs on the entire Anton 2 machine, utilizing all its 128 custom ASIC nodes.
| Chemical system (PDB ID) | Number of atoms | Approximate performance (microseconds/machine-day)* |
|---|---|---|
| DHFR (5DFR) | 23,558 | 61.3 |
| aSFP (1SFP) | 48,423 | 53.0 |
| FtsZ (1FSZ) | 98,236 | 26.0 |
| T7Lig (1A01) | 116,650 | 21.9 |
| bILAP (1BPM) | 132,362 | 18.7 |
| f1atpase | 327,506 | 7.9 |
| Tiled FDH-H** | 700,184 | 3.6 |
* All simulations used 2.5-femtosecond time steps with long-range interactions evaluated at every other time step and a Nose-Hoover thermostat applied every 100 time steps. Performance was measured on a 128-node Anton 2 machine like the one hosted by PSC. Simulation performance in microseconds/machine-day is approximately 50% higher if 4fs time steps and Hydrogen Mass Repartitioning are used (see Reference 3). ** This system represents eight copies of FDH-H (formate dehydrogenase H) in a single 190 Angstrom cubical box.
References
- Anton 2: Raising the Bar for Performance and Programmability in a Special- Purpose Molecular Dynamics Supercomputer, D. E. Shaw et al., Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis (SC14), (2014)
- Accurate and Efficient Integration for Molecular Dynamics Simulations at Constant Temperature and Pressure, R. A. Lippert et al., Journal of Chemical Physics, vol. 139, no. 16, 2013, pp. 164106:1–11.
- Desmond/GPU Performance as of October 2015, M. Bergdorf et al., D. E. Shaw Research Technical Report DESRES/TR—2015-01, 2015 (available at http://www.deshawresearch.com/publications.html)