More than a year ago, computational scientists at the Department of Energy’s Oak Ridge National Laboratory that raised a serious question about a long-standing methodology used by researchers who conduct molecular dynamics simulations involving water. What if using the standard 2 femtosecond (2 quadrillionths of a second) time step — the time interval at which computer simulations are analyzed — leads to inaccurate results?
Now, the same ORNL team has that reaffirms their original observations by showing how using these “standard” time steps can affect simulations of liquid water. The team’s calculations reveal that the potential for errors caused by using a 2 (or more) femtosecond time step is even greater than they had anticipated.
“I was a little bit surprised. I was hoping for much more subdued effects, but the errors can be big,” said co-author Dilip Asthagiri, a senior computational biomedical scientist in ORNL’s Advanced Computing for Life Sciences and Engineering group. “What we are saying is, ‘With the benefit of knowledge gained over the last 50 years of studies on water, let’s do the statistical mechanics at the most accurate, converged level so that we can better assess the errors in the simulation and focus on issues that we should address.’ It’s an evolution of science.”
Previous research
Water is the most prevalent component of biomolecular simulations — from protein ensembles to nucleic acids — and inaccurately simulating it can lead to errors for research results in biomolecular structure, function, dynamics and assembly. Industries ranging from pharmaceutical to petroleum rely on accurate water simulations to attain a competitive edge for their products and operations.
, using anything greater than a 0.5 femtosecond time step can lead to violations of equipartition — the requirement for simulations that the average kinetic energy for each type of motion should be the same. This lack of equipartition can introduce errors in both dynamics and thermodynamics when simulating water using a rigid-body description.
Treating water as a rigid body rather than as a flexible bond between hydrogen and oxygen allows scientists to use longer time steps. The technique dates to 1977, when complex computations were more time-consuming and expensive. The longer the time step, the greater the total physical time that can be modeled in the simulation. Although the Oak Ridge Leadership Computing Facility’s Frontier and other modern supercomputers can reduce the time to solution enough to enable fewer approximations, using a smaller time step inevitably increases the computational cost.
“Sometimes with this kind of study, people may not know how to deal with the conclusions. Eventually, people become aware of it, and then it gradually gets adopted down the road,” said co-author Tom Beck, section head of Science Engagement in the National Center for Computational Sciences at ORNL. “But the point is to say, ‘If we’re going to really do predictive science for a given model — in order to test that model versus experiment — we need to accurately represent the underlying thermodynamics and dynamics.’”
New findings
In their new study, the team — Asthagiri, Beck and Arjun Valiya Parambathu — used Frontier to simulate samples of liquid water, and they explored various system sizes and different combinations of temperature and pressure. They conducted these simulations at time steps that ranged from 0.5 to 3.5 femtoseconds in intervals of 0.5 femtoseconds.
“One of the issues in capturing the physics of water is to accurately capture the pressure and volume behavior,” Asthagiri said. “What our study shows is, using the same pressure, doing the simulation at longer time steps will give you different volumes or give you alternatively different densities. But if you go to very short time steps, the results all converge, and you get a consistent prediction.”
The ORNL team has also been investigating the role of hydration in the thermodynamics of protein folding. Understanding protein folding is a crucial area of study in biology, particularly for research into the molecular basis of diseases and drug discovery. The were researchers in protein structure prediction and computational protein design whose work was assisted by ORNL’s High Flux Isotope Reactor. An important physical effect in protein folding is the change in the volume of the protein-water system.
“What we find is that in the simulation of neat liquid water, the error in total system volume that one makes in using a long time step can be as high as or higher than the typical volume change in protein folding. The same goes for the hydration free energy. While the implications of this finding will need to be worked out for actual protein folding and assembly processes in liquid water, our results suggest the need for much care,” Asthagiri said.
Although the team did hear from some skeptical peers about the findings of their initial study, it was also cited as a reference in papers published in Molecular Physics, Physical Chemistry Chemical Physics, ACS Nano and the Journal of Chemical Theory and Computation.
“Why are we even doing this? Our eventual goal is going to much larger simulations of biological molecules and of cellular systems, and that’s really what we want to go and study,” Asthagiri said. “But we’ve been sidetracked into this because we just want to get the basics right before we go and do these extremely large simulations. We need to know how to simulate the matrix of life better so that we can study the biological processes better.”
ORNL houses the Frontier supercomputer at the OLCF, a DOE Office of Science user facility.
UT-Battelle manages ORNL for DOE’s Office of Science, the single largest supporter of basic research in the physical sciences in the United States. DOE’s Office of Science is working to address some of the most pressing challenges of our time. For more information, visit . — Coury Turczyn
