Fixing and flexing biomolecular force fields
03.20.14 - Permalink
by Trish Barker
A University of Utah research group is using the massive scale of Blue Waters to rapidly and rigorously evaluate the force fields used in molecular dynamics simulations.
Thomas Cheatham’s University of Utah research team uses molecular dynamics simulations to better understand biomolecules, like nucleic acids and proteins. Improving and validating their methods, and the Amber tools and force fields that they use, is a critical part of that work.
“You can model anything; that doesn’t mean it’s real,” points out Cheatham, an associate professor of medicinal chemistry. “We constantly have to assess and validate results with comparison to experiment.”
In the Amber molecular dynamics software, force fields define all of the ways atoms can interact: bonds, angles, bond rotations, electrostatic and van der Waals' forces. Changing the value assigned to one of these fields will change the outcome of the simulation.
“The force field is the energy function we use to describe the system; it’s essentially the recipe for how this molecule moves and reacts to its environment,” explains Christina Bergonzo, a postdoctoral fellow in Cheatham’s research group.
Getting this force field right—meaning that the simulations match what has been observed experimentally—is tricky, particularly for RNA, which is flexible and adopts many different structures, or conformations.
“What we often find,” Bergonzo explains, “is that we run simulations and our force fields start to break down,” generating results that diverge from experimental observations. “So we make fixes to the force field, and run it a little bit longer, and it falls apart again.”
After a force field breaks in this way, researchers propose tweaks, but “a fix may only be as good as the amount of time you have run a simulation,” Bergonzo says.
But with the massive scale of NCSA’s Blue Waters supercomputer—particularly with the system’s GPU nodes—the Cheatham research team can rapidly run more extensive Amber simulations to more rigorously evaluate a variety of force fields.
“If it doesn’t fall apart across a microsecond, someone might call that OK. But we can run it and run it and run it,” Bergonzo says. “This machine really helps to eliminate the issue of limited sampling.”
“We’re getting about 100 nanoseconds per day per GPU,” Cheatham says. “When combined with technologies that couple together ensembles of independent simulations for faster sampling, these simulations quickly aggregate to the biorelevant 10-100 microsecond time scale and beyond.” This enables the group to fully converge the structural ensemble of DNA helices or RNA tetranucleotides, and hopefully soon larger RNA motifs.
In work recently published in The Journal of Chemical Theory and Computation, the group ran several variations of replica exchange molecular dynamics on Blue Waters, testing their ability to generate the range of conformations that have been observed for an RNA tetranucleotide. They determined that multidimensional REMD (M-REMD) generated the slate of possibilities more efficiently than temperature REMD or Hamiltonian REMD alone.
“The results tell us that…our force field is not reproducing experiment,” Bergonzo says. “And because we did this type of multidimensional replica exchange, we can pinpoint the ways in which we aren’t reproducing experiment. Our simulations prefer these globular and over-stacked structures.” The next step will be to figure out what force field adjustments will bring the simulations in line with experimental observations, and then apply better force fields to study larger conformational motifs to not only further validate and assess the approach, but also to provide novel biomolecular insight into RNA structure, dynamics, and function.
Generating the converged ensemble of structures in this way “becomes like a benchmark for what we think people should be doing in order to test current force fields and to prepare new force fields,” Bergonzo says. “We’re really excited that multi-dimensional replica exchange is kind of going to be our lab’s benchmark way of testing force fields.”
“We’re interested in advancing these methods and understanding their limitations, so people who are applying them can figure out the best way to do that. And Blue Waters is enabling us to do that significantly faster than we could on alternative resources,” Cheatham says.
For more information
Bergonzo, C., Cheatham, T. E., III., Henriksen, N. M., Roe, D. R., Roitberg, A. E., & Swails, J. M. (2013). Multidimensional Replica Exchange Molecular Dynamics Yields a Converged Ensemble of an RNA Tetranucleotide. J. Chem. Theory Comput., DOI: 10.1021/ct400862k
Blue Waters is supported by the National Science Foundation through awards ACI-0725070 and ACI-1238993.