Project Ideas

TPS (Transition Path Sampling) on Grid Distributed Computation and Storage Resources

TPS is accomplished by generating multiple trajectories from a potential midpoint in a desired path between two molecular structures (conformations). Large numbers of trajectories must be attempted to generated a suitably sized set of successful trajectories which can be refined to indicate an optimal transition path. Generation of this large set of trajectories requires computation and storage resources on a scale best provided via developing grid technologies.

REM (Replica Exchange Method) Acceleration Tools

The REM method (also known as parallel tempering) is a popular method to accelerate sampling of systems prone to local trapping by rough energy landscapes. Multiple papers have reported implementations and modifications of REM however there is no generic simulation tool (program) which allows modular algorithm development and efficient simulation of large systems. This project would restructure the existing REM code in the ProtoMol framework and/or external Condor REM application to allow for modular development of REM simulation algorithms.

Docking on Grid Distributed Computation and Storage Resources

There are currently multiple competitive protein-protein and protein-ligand docking algorithms. Most of which utilize a set of protein/ligand candidates (or a geometric/energetic average) and then numerically compute the energetic feasibility of the dock. The computation and storage requirements grow rapidly with increases in the set and molecular size. Harnessing grid distributed resources has the potential to greatly expand capabilities.

Parallel and Distributed MDL (Molecular Dynamics Language)

Molecular simulation is both computation and storage intensive, prompting the development of distributed algorithms. The MDL prototyping language is currently only designed for single processor functionality. Providing mechanisms for the distributed utilization of MDL will allow both single cpu and multiple cpu developers a single prototyping tool. More Info...

Reaction Path Calculation (ProtoMol, MOIL, or Python/MDL)

Most molecular dynamics simulations are initial value problems; however when one has knowledge of initial and terminal configurations a boundary value approach can be employed. Although the computation required to solve an optimal boundary value problem is often much larger than an initial boundary trajectory, the boundary value problem can be solved approximately using a very large time step, thus making the computation cost an arbitrary variable with respect to the accuracy desired. Reviewing the MOIL implementation and adding an enhanced implementation to ProtoMol or MDL would be of great resource to algorithm developers.

Free Energy ABF (Adaptive Biasing Force) in ProtoMol

The adaptive biasing force method is an efficient technique to compute the potential of mean force along a reaction coordinate and for alchemical transformations. The implementation of this method in ProtoMol will allow for its incorporation into additional integration and sampling schemes while providing a stable code framework for its further development.

Parallel MultiGrid at the Pittsburgh Super Computing Center

In N-body problems, such as weather forecasting, galaxy formation, and protein folding, force calculations dominate CPU usage. The exact mathematical solution generally scales as O(N^2). Several methods, particle mesh Ewald (PME) for example, asymptotically improve this bound through approximations, but are often hard to parallelize. Recently, the multigrid summation (MG) method, which scales linearly with system size, was extended to run in parallel {Izaguirre et al. 2005}. This project proposes the use of advanced parallel computers to demonstrate the scalability of MG. To this end, we wish to increase the size of the test cases up to 400,000 atoms and to run simulations on computers with up to 1024 nodes at the Pittsburgh Super Computing Center. Note: this project is dependent on PSSC approval

APBS (Adaptive Poisson-Boltzmann Solver) in ProtoMol and integrated with PMV

The computational cost of molecular simulation is severely impacted by the addition of solvents which are critical to representative molecular behavior. The number of atoms introduced by an encapsulating box of water molecules can increase the total number of atoms by an order of magnitude. Multiple implicit solvent methods have been introduced to address this computational limitation. The APBS method for implicit solvation is a promising method for which integration with ProtoMol would enhance it evaluation and reduce the computational complexity of large and or coarsened molecular systems.

Simulation and Analysis of GNM/ANM (Gaussian/Anisotropic Network Models)

One way to address the 6N dimensional phase space of large molecules is to reduce N by coarsening the molecular model. The GNM and ANM methods have proven successful in initial tests. Further testing in line with experimental NMR data and implicit solvents has the potential to increase the total resolved simulation timescale by orders of magnitude.