Using Molecular Dynamics for Model Generation

Goals

The goal of this part of the project was to generate trajectories for the molecules that are the subject of our interactive visualization demonstration. A trajectory can be thought of as a series of snapshots of the positions of each atom in the molecule taken over time. The positions for each atom were updated by a molecular dynamics simulation that attempts to model the forces acting on the molecule in its physical environment. Our intention was to build molecular animations that correspond to movements that are physically realistic or at least plausible.

Tools

I primarily used two tools for this part of the project. They are both developed by and available from the Theoretical Biophysics Group at the University of Illinois at Urbana-Champaign.

Resources

Introductory-level information on molecular dynamics simulations is available in many places on the web. Two of the resources that I found most helpful were:

Running the Simulations

In an attempt to get interesting, yet plausible, behaviors for the molecules in the simulations without having the requisite background in organic chemistry and molecular biology, I arrived at four possibilities:
  1. A fallback position of simulating the heating of one of our molecules from absolute zero to room temperature. This would show only the vibration of the atoms due to thermal energy, but it is quite realistic.
  2. Simulate a superheating of one of our proposed molecules and watch for possible configuration changes due to the extra energy or at least the breaking of bonds.
  3. Manually reconfigure one of our molecules to a more "open" configuration and simulate the molecule at room temperature to see if it will fold back to its original configuration as defined by the PDB file available from the Protein Data Bank.
  4. Build a realistic environment for one of our target molecules to perform its biological function.

Process

The steps in the following process are fairly standard for running molecular dynamics simulations. This is how I generated results using NAMD and its associated tools. The NAMD users guide has an example called the BPTI example that is a good place for a step by step walk through of this procedure.
  1. Prepare the PDB description of the system to be simulated.

    This step included manually editing the PDB files to do things like removing water molecules and free ions. This is also the step where the modifications were made to slightly unfold the protein using the scripting capabilities of VMD. I wrote a tcl script to walk the backbone of the currently active protein and open all the angles between residues to give us a new starting configuration for the simulations.

  2. Run the psfgen utility to build an X-PLOR format .psf (protein structure file) file that is needed for input to NAMD.

    This step required writing a psfgen configuration file and used input in the form of a topology file that describes the common amino and nucleic acids in proteins. The topology file is shipped with CHARMM and is available in various locations on the web. The documentation with NAMD includes an example and a location for retrieving this topology parameter file. This is also the place where the water molecules and ions removed in step one could be added back in from a secondary pdb file as a separate segment.

  3. Perform energy minimization and equilibration of the system in NAMD.
  4. Run the dynamics simulation of the system in NAMD.

    These last two steps are combined automatically by NAMD. The parameters for both steps are specified in a NAMD configuration file. NAMD uses a force field parameter file (also shipped with CHARMM) that provides parameters for all the inter-atom forces it simulates. In the example, you can see parameters that indicate how many timesteps of minimization should be done and how long the simulation itself should last. There are also parameters that allow the temperature to be periodically rescaled or changed by a specified amount. I relied heavily on this feature to generate the trajectories.

  5. View the results in VMD to evaluate them with respect to our "interesting" criteria.
These steps were iterated until either interesting results were obtained or it was clear that the approach wasn't going to work.

Results

Here's what came of each of the four possibilities I explored:
  1. Our fallback position of animating the molecule in its room temperature state worked like a charm. I was able to generate trajectories for any of the molecules we wanted under these conditions. We settled on the anthrax toxin (PDB id: 1ACC) as our final target molecule and that data made it into the final demonstration.

  2. The idea of superheating a molecule to watch parts of it "evaporate" away didn't work out so well, because the simulation always seems to become unstable before the bonds broke as the energy in the system was increased.

  3. The idea of "opening" a molecule to watch it refold showed promise, but was abandoned due to time constraints. As mentioned above, I wrote a tcl script that could be run in VMD to reconfigure the molecule. Using that script I obtained reasonable results for a small molecule (PDB id: 1CRN). Those results are available as a pair of gzipped DCD and PDB files that (once unzipped) can be viewed interactively in VMD. Note that the DCD format is binary and may not be portable between big and little endian machines. VMD is supposed to take care of this above version 1.4.

    The script produced very undesirable results for larger, more complex proteins because it tended to move too many atoms close enough together to bond. The simulation and viewing software then created bonds between these atoms in close proximity that are not physically possible. With more work, I think the script could be modified (or rewritten) to avoid changing angles that result in these "impossible" bonds being created.

    The results would also be more realistic if we had been able to do the simulation in a solvent (water). To do this required building a pdb file of a suitable sized cube of water that is in equilibrium and cutting out a hole in the center to merge in the protein. I imagine this is not difficult once you've done it a couple of times, but it was prohibitively complicated given our time frame and my experience level.

  4. The fourth idea of creating a realistic environment for one of our molecules was abandoned rapidly after seeing the difficulty involved just in getting a realistic solvation of the molecule. In addition to that problem, it would have required researching in detail the biochemical processes in which our molecules participate. No one in the group seemed to have the background necessary to get this done in a reasonable amount of time.

Lessons Learned

I probably learned more about molecular dynamics from (my portion of) this project than I did about visualizing the results since my output was a sequence of PDB files that others in the group used to generate models. The following is a short list of some of what I've learned: