Our research is very computer intensive. In our lab, we have a Silicon Graphics Indigo R4000.
We have run a major portion of our simulations at the Pittsburgh
Supercomputing Center on the Cray T3D (left) and the Cray T3E. Both
machines have 512 processors for massively parallel computing. Currently,
we have begun work at NCSA (National
Center for Supercomputing Applications) using the Power
Challenge Array. And finally some work is done using MU's finest number
cruncher, SHIVA.
SHIVA (right) is a Silicon Graphics Power Challenge L with 8 R10000 processors
for parallel computing.
This computer is managed by Dr.
Hossein Tahani and others at Research Computing Services.
Our research is focusing on how a peptide that exhibits little or no secondary structure (i.e. alpha-helix and beta turns for instance) in water interacts with a membrane like interface and ultimately forms a secondary structure. This interface is a boundary where a polar aqueous phase and a hydrophobic non-polar phase meet.
Our simulations fall into the category of Molecular Dynamics (MD). In MD Newton's equations of motion are solved given an emperically based force field. MD allows us to view how the molecular system evolves through time and to constuct average properties of the system given a long enough simulation. There are many issues to consider in MD -- long range interactions, polarizability, suitability of the force field, different ensembles, etc. that can make this area of computational chemistry especially challenging. While it would be desirable to use the best of all possible methods (even use electronic structure theory) and simulate for very long times (perhaps a whole second), many times we are limited in simulations of biomolecular assemblies by the size of the system and the computational resources (among other things).
Given a starting position for all the atoms in the system; a minimization of the energy is carried out. Minimization includes first defining the force field parameters for the atoms. The parameters include force constants for bond stretching, bond bending and torsional rotation. Also, the electrostatic and van der Waals (vdW) interactions make up the non bonded forces. Electrostatic interactions are generally just point charges located at the center of the atom that are calculated by Coulomb's law. MD usually contains no calculation of dipole-dipole interactions; though the charges and hence dipoles of individual molecules are enhanced to implicitly include polarization effects. The minimization occurs usually by second derivative methods in order to reduce the gradient of force on all atoms to an acceptable level.
Then the molecular dynamics is carried out by first heating the system to the experimental temperature by scaling of the velocities. How the system evolves through time is specified by the force field and an integration time step that determines where the atoms should be at a later time. Molecular Dynamics (MD) requires the use of a very small timestep, typically 1 femtosecond, to acheive accurate results. Hence this limits the timescale of the simulation and what is possible to observe and calculate. Once the experimental temperature is reached, then typically some equilibration is carried out followed by the production run. The production run should be long enough to sample the property of interest and to give good statistics.
Our first membrane like environment (membrane mimetic) is a two phase system made up of TIP3P water molecules and carbon tetrachloride molecules (left). This two-phase system mimics the polar and non-polar sections of a cell membrane. This simple mimic should be able to reproduce the hydrophobic driving force for peptide insertion into the cell membrane. This mimic though leaves out the effect of the headgroups on the peptides and the interface between polar/non-polar phases is much sharper. The reason for the simpler model is due to computational cost. Our biphasic system that is 42 angstoms cubed consists of approx. 5000 atoms compared to some other membrane mimics that contain 2-3 times more atoms. Also, the peptides in the biphasic cell are able to reorient if placed in an unfavourable position and orientation on a faster time scale. These simulations enable us to contruct good starting models for simulation in the explicit micelle or lipid bilayer.
This membrane mimetic is simulated with the MD program CHARMM and Sybyl from Tripos, Inc. Modelling how peptides interact with this interface is done by placing the peptide in the biphasic cell. Since the folding process can take up to nanoseconds or more, this is very expensive to simulate given a 1 femtosecond time step. Therefore we place the peptide at the interface in a conformation that agrees with experiments done where the interface is a micelle and see how it evolves from there. The phenomena of interest will be to determine if this system is sufficient for a membrane mimetic and to determine the forces that stabilize the secondary structure of the peptide compared with the random coil conformation in water.
Our second membrane mimetic is an sodium dodecylsulfate (SDS) micelle. This system is a more realistic mimetic and results of the simulation can be more directly tied to experiment but at a computational cost. The micelle is a spherical aggregate of 60 lipid molecules. Also included are 60 sodium ions (Na+) to have a neutral system. Finally the system is solvated with TIP3P water to give a total of 15,776 atoms. Modelling this system alone is a challenge, much less this with a peptide. These simulations will be carried out with the program CHARMM. This program has a lipid and protein parameter set that have been shown to reproduce many experimental parameters.
CHARMM
example input files for minimization, dynamics and analysis
Biomolecular
Computational Chemists and Programs
NMR
Investigations of Peptides with Membrane Mimics
How
to parallelize FORTRAN code on the SGI Power Challenge
