Inside the Laboratory: How Computational Approaches Can Improve Understanding of Molecular Behavior

Inside the Laboratory is a joint series with LCGC International and Spectroscopy, profiling analytical scientists and their research groups at universities all over the world. This series spotlights the current chromatographic and spectroscopic research their group is conducting, and the importance of their research in analytical chemistry and specific industries. In Part 2 of this “Inside the Laboratory,” feature on George Shields, a professor of chemistry at Furman University and the founder and director of the Molecular Education and Research Consortium in Undergraduate Computational ChemistRY (MERCURY), Consortium, we discuss his research into computational approaches to improve our understanding of molecular behavior in both biochemistry and atmospheric chemistry and his work applying replica exchange molecular dynamics (REMD) for breast cancer drug design (1,2).

George Shields is a professor of chemistry at Furman University and the founder and director of the Molecular Education and Research Consortium in Undergraduate Computational ChemistRY (MERCURY). | Photo Credit: George Shields.

Your research integrates quantum chemistry, Monte Carlo, and molecular dynamics methods—can you explain how combining these computational approaches enhances your understanding of molecular behavior in both biochemistry and atmospheric chemistry?

Science is an interdisciplinary endeavor, and sometimes experimentalists and theorists work together to understand phenomena, and other times one scientist uses multiple methods in their own group to learn about the natural world. In my laboratory, we find that multiple methods are required to make sure we have a high degree of confidence in our results. So, for instance, we might use Monte Carlo or molecular dynamics techniques in order to explore a very large number of molecular complexes that could be formed in the atmosphere. Imagine millions of clusters of sulfuric acid, ammonia, and a handful of water molecules. It is not possible to do high level quantum chemistry on a million molecular clusters, but we can use a simpler, faster theory to reduce a million to a few hundred, and then a high level theory to reduce the number down below a hundred, giving us a higher degree of confidence that we have predicted the actual structures of the complex of interest that would be formed in our atmosphere. For biochemistry applications, the molecules are so large that we can’t use quantum chemistry, so we use molecular dynamics to learn about the system of interest. Since we understand the error bars associated with different methods, it allows us to have a better feel for the uncertainties associated with results from a molecular dynamics simulation versus a quantum mechanical simulation.

You've applied replica exchange molecular dynamics (REMD) for breast cancer drug design, particularly with AFP-derived peptides—what makes REMD a uniquely powerful tool in this context, and how did it lead to the identification of conserved reverse beta-turn conformations?

REMD is a very powerful technique, because it allows the system to climb over potential energy barriers. It does this by running multiple simulations of a system at different temperatures.At high temperatures the energy from heating boosts the system above a potential energy barrier, so that different conformations can be found. For this project, my students had been using molecular dynamics to understand the conformations of AFP-derived peptides. A senior in my laboratory, Katrina Lexa, presented her work at the Sanibel Symposium in February of her senior year, and at the conference Carlos Simmerling at Stony Brook University and Adrian Roitberg at University of Florida told Kat about REMD, and Karl Kirschner and Kat got it working after the meeting and we discovered that only certain peptides with a particular sequence had a common structure. When we told our collaborators at SUNY Albany Medical School of our results and they made and tested the peptides, they found that our predictions inhibited breast cancer in their mouse and rat systems, which led to these two patents. Only structures with a proline residue in the second position of a particular peptide sequence make the conserved reverse beta-turn. Kat went on to graduate school at the University of Michigan and worked at Merck and various startup companies and is now Senior Director at Tenvie Therapeutics.

How do you design computational projects to be both accessible to undergraduates and meaningful within the broader goals of your NIH, DOD, and NSF-funded studies?

Designing meaningful computational projects for undergraduates, where you can finish a project without being scooped by a research university which may have an army of graduate students and postdocs is an art form. It is made more difficult by the fact that in order to obtain money from the federal government your proposed science must pass a very high bar. I feel like because my background was in chemical physics, were I applied mass spectrometry (MS) and semi-empirical methods to understand ion-molecule reactions in Tom Moran’s laboratory for my PhD at Georgia Tech, and my postdoctoral research was in structural biology, where I learned how to determine the crystal structures of protein-DNA complexes using X-ray crystallography and computational techniques in Tom Steitz’s lab at Yale, that I was well-prepared to pick problems that were both interesting and possible for undergraduates to make progress on. I taught myself computational chemistry during my first position as an Assistant/Associate Professor at Lake Forest College (1989–1998) and have used computational chemistry to conduct research in a wide range of fields. I would say that about half of my projects have led to publications, and all of them have enhanced the understanding of science for the undergraduates who have worked on these projects. When you are a generalist, then you notice some things in the literature where you might be able to make a difference, which leads to a new area of research.

In your study on computational pKa prediction, solvation plays a critical role. What have you found to be the most reliable solvation models or techniques for accurately predicting deprotonation in aqueous environments?

The pKa prediction is an example of my group picking up this work at the right time. I was working on a collaboration with Don Landry, a physician and scientist at Columbia College of Physicians and Surgeons, where we were trying to design molecules that could be used by his group to generate catalytic antibodies that would destroy cocaine. Ed Sherer and Gordon Turner (now directors at Merck and Novartis, respectively) had done some great work on this project while I was at Lake Forest College. Don asked me one day if we could predict the charged state of molecules he was trying to synthesize, and I said sure. My group looked into this when I moved to Hamilton College. When I read the literature, a future Nobel laureate had written a paper where he explained that the solvation models were simply not reliable enough to be used to predict pKa values. Ed was working on his PhD with Chris Cramer and Chris and I had been talking about computational chemistry and solvation models for several years, and I knew that Chris and Don Truhlar, both at the University of Minnesota, had been working hard on better solvation models. When Matt Liptak joined my lab after his first year at Hamilton, he figured out the right combination of gas-phase and solvation-phase models to accurately calculate the pKa of carboxylic acids and substituted phenols, which led to two JACS publications. Matt went on to obtain a Ph.D. at the University of Wisconsin and a postdoc at the University of Rochester, and he has been a professor of inorganic chemistry at the University of Vermont for quite a while now. Once we showed that accurate pKa calculations were possible, the field exploded, and we were no longer to keep up with research universities. Matt’s two JACS publications have been cited 542 and 543 times to date, according to Web of Science. His eight publications on this project, many with other Hamilton College undergraduate have been cited a total of 1694 times by WOS. The most reliable solvation model today is the SMD model of Cramer and Truhlar, which I’ve noted in a couple of review articles I’ve written.

References

1. Mercury Consortium, Conference 2025. Mercury Consortium. Available at: https://www.mercuryconsortium.org/conference-2025/(accessed 2025-08-05).
2. Furman University, George C. Shields. Furman.edu. Available at: https://www.furman.edu/people/george-c-shields/ (accessed 2025-08-05).

Definitions of U.S. Organizations

NIH (National Institutes of Health), DoD (Department of Defense), American Chemical Society Petroleum Research Fund (ACS PRF), and NSF (National Science Foundation)