HPC Amber 12 New Features

A number of features not available in AMBER 11 have been implemented in AMBER 12. The new features are described below, along with the Sander functionality previously unavailable in PMEMD and the GPU Accelerated PMEMD.

Updated Features in AMBER 12

  • Force fields: a new fixed-charge protein force field, ff12SB, enchanced support for polarizable potentials and a new modular lipid force field Lipid11 designed to be compatible with the other pairwise additive AMBER force fields
  • Expanded options for numerical Poisson-Boltzmann solvation calculations, including models for membrane systems and support for periodic systems
  • An enchanced 3D-RISM integral equation model, using the Kovalenko-Hirata (and other) closure approximations, with a better treatement of aqueous electrolytes
  • Improved ideas for self-guided Langevin dynamics and accelerated molecular dynamics, to enchance sampling along soft degrees of freedom
  • Semi-empirical quantum calculations can use d-orbitals, allowing the use of Hamiltonian models such as AM1/d and PM6
  • QM/MM calculations can interface with a variety of external quantum chemistry programs, expanding the types of quantum models available
  • Additional features from Sander have been added to the pmemd code for both CPUs and GPUs, including Temperature Replica Exchange, Isotropic Periodic Sum, Accelerated Molecular Dynamics and support for various harmonic restraints based on the use of NMRopt on GPUs
  • Expanded methods are available for free energy calculations that change Hamiltonian models, including better procedures for appearing and disappearing atoms, and tighter integration with replica-exchange simulations
  • New facilities are present for using electron density maps (e.g. from cryo EM/ET experiments) as constraints, and to support rigid (or partially flexible) groups in simulations

PMEMD Functionality Available in AMBER 12

Particle Mesh Ewald (PME) Molecular Dynamics (PMEMD) is a reimplementation of a subset of Sander functionality that has been written with the major goal of improving the performance of the most frequently used Sander methods for long time scale simulations of explicitly or implicitly solvated systems. PMEMD supports PME, Generalized Born (GB), and Analytical Linearized Poisson-Boltzmann (ALPB) simulations using both the AMBER and CHARMM Force fields. The AMOEBA polarizable force field, is also supported but via a separate PMEMD executable, pmemd.amba. One of the major additions to PMEMD is the support for acceleration of both PME and GB calculations using NVIDIA GPUs.

The following new functionality is available in PMEMD in Amber 12:

  • multi-pmemd (which uses the same flags as multisander)
  • replica exchange MD simulations (REMD) in both Temperature space and Hamiltonian space
  • restraintmask and bellymask atom mask specification options enabled
  • IPS simulations
  • Generalized Born model enabled (igb = 8)
  • LCPO surface area potential terms enabled for GB simulations (gbsa = 1)

New Functionality for GPU Accelerated PMEMD in AMBER 12

A number of enhancements have been made to the GPU accelerated code in AMBER 12. These enhancements include:

  • temperature replica exchange,
  • umbrella sampling (through nmropt = 1),
  • simulated annealing (through nmropt = 1),
  • all GB models,
  • Jarzysnki sampling.
  • isotropic periodic sum,
  • support for extra points,
  • improvements to the way in which specific GPUs are selected when running calculations,
  • support for parallel runs, and
  • accelerated MD

Sander Functionality Not in GPU Accelerated PMEMD

The GPU accelerated version of PMEMD supports both explicit solvent PME simulations in all three canonical ensembles (NVE, NVT and NPT) and implicit solvent GB simulations. However, the following Sander functionalities are currently not supported in GPU Accelerated PMEMD in AMBER 12.

  • ibelly /= 0
    simulations using belly style constraints
  • if (igb /= 0 and cut < systemsize) GPU accelerated implicit solvent GB simulations do not support a cutoff.
  • nmropt > 1
    In addition, for nmropt = 1 only features that do not change the underlying force field parameters are supported. For example, umbrella sampling restraints are supported as is Jarzynski sampling, as well as simulated annealing functions such as variation of Temp0 with simulation step. However, varying the VDW parameters with step is not supported.
  • nrespa /= 1
    multiple time stepping
  • vlimit /= -1
    for performance reasons the vlimit function is not implemented on GPUs
  • numextra > 0 with MPI
    extra points are currently only supported for single GPU runs
  • imin = 1 with MPI
    minimization is currently only supported for single GPU runs
  • es_cutoff /= vdw_
    cutoff Independent cutoffs for electrostatics and van der Waals
  • order > 4
    PME interpolation orders of greater than 4.
    Additionally, there are some minor differences in the PME output format. For example, the Ewald error estimate is not calculated when running on GPUs. It is recommended that you first run a short simulation using the CPU code to check the Ewald error estimate is reasonable and that the system is stable.

Sander Functionality Not Implemented in PMEMD

Since PMEMD is an incomplete implementation of Sander, certain options available in Sander are not available in PMEMD. The following is a list of Sander functionalities that have not been implemented in PMEMD.

Options in the input (&cntrl) file that are not supported:

  • imin = 5
    trajectory analysis is not supported.
  • nmropt = 2
    a variety of NMR-specific options such as NOESY restraints, chemical shift restraints, pseudocontact restraints, and direct dipolar coupling restraints are not supported.
  • idecomp! = 0
    energy decomposition options used in conjunction with mm_pbsa
  • ipol! = 0
    polarizable force field simulations, other than amoeba, which is supported in pmemd.amba
  • igb = 10
    Poisson-Boltzmann simulations
  • igb = 6 or 8
  • ntmin > 2
    XMIN and LMOD minimization methods
  • isgld! = 0
    self-guided Langevin dynamics
  • noshakemask
    the noshakemask string option
  • icfe! = 0
    calculation of free energies via thermodynamic integration
  • itgtmd! = 0
    targeted molecular dynamics
  • ievb! = 0
    Empirical Valence Bond methods.
  • ifqnt! = 0
    QM/MM methods
  • icnstph/ = 0
    constant pH simulations
  • ineb! = 0
    nudged elastic band (NEB) calculations (these calculations are done by sander.MPI)
  • iamoeba!=0
    the amoeba polarizable potentials of Ren and Ponder are not supported in PMEMD, but are supported in pmemd.amba

Methods that are not supported:

  • Solvent Caps
    solvent cap simulations
  • LES
    the Locally Enhanced Sampling method
  • REM = 2
    the partial REMD method (for LES)

Namelist options that are not supported:

  • &debugf namelist
    use of the &debugf namelist is only supported in a very limited way — specifically, only the do_charmm_dump_gold option is supported

&ewald options that are supported, but only with the indicated default values:

  • ew_type = 0
    only Particle Mesh Ewald calculations are supported. ew_type = 1 (regular Ewald calculations) must be done in sander.
  • nbflag = 1
    the nbflag option is ignored for MD and all nonbonded list updates are scheduled based on "skin" checks. This is more reliable and has little cost. The variable nsnb still can be set and has an influence on minimizations. For PME calculations, list building may also be scheduled based on heuristics to suit load balancing requirements in multiprocessor runs.
  • nbtell = 0
    the nbtell option is not particularly useful and is ignored.
  • eedmeth = 1
    only a cubic spline switch function (eedmeth = 1) for the direct sumCoulomb interaction is supported. This is the default, and most widely used setting for eedmeth. On some machine architectures, energies and forces are actually splined as a function of r**2 to a higher precision than the cubic spline switch. One consequence of only supporting eedmeth 1 is that vacuum simulations cannot be done (though generalized Born nonperiodic simulations are available).
  • column_fft = 0
    this is a sander-specific performance optimization option. PMEMD uses different mechanisms to enhance performance, and ignores this option. It is suggested that new PMEMD users simply take an existing sander mdin file and attempt a short 10-30 step run. The output will indicate whether or not PMEMD will handle the particular problem at hand for all the functionality that is supported by "standard" sander. For functionality that requires special builds of sander or sander-derived executables (LES), there may be failures in namelist parsing.

859-218-HELP (859-218-4357) 218help@uky.edu