Decades later, a Yale chemist’s water simulations continue to make waves

One of the most influential science studies of all time started with a modest minicomputer, some simulated boxes of water molecules, and a grand vision for computer-aided chemistry.

The year was 1982. William Jorgensen — now a revered, Sterling Professor of Chemistry in Yale’s Faculty of Arts and Sciences — was a 32-year-old assistant professor at Purdue University in West Lafayette, Indiana.

At the time, Jorgensen had already helped to develop the CAMEO computer program for predicting the products of chemical reactions (While at Purdue he’d also learned from 1979 Nobel Prize laureate Herbert C. Brown how to hold a piece of chalk so it wouldn’t squeak on a blackboard).

But now he had an idea that would revolutionize drug discovery worldwide. 

Although his research background was in quantum chemistry with a dash of organic work, he had become enamored with a notion more in the realm of statistical mechanics: What if he could use a “minicomputer” — which, in those days, was the size of a tall cabinet — to create accurate models showing how water molecules interact? And, if successful, could he simulate molecules in solution starting with simple liquids and progressing to biomolecular systems?

The resulting study, “Comparison of simple potential functions for simulating liquid water,” published a year later in the Journal of Chemical Physics, would go on to dominate the field of modeling liquid water, with a trail of related work that continues into the age of machine learning and supercomputers. 

As of this month, more than 45,000 other studies have cited Jorgensen’s landmark work. Earlier this year, it ranked 88th in a Nature list of most-cited studies of all time (reflecting a database going back to the year 1900). It is not a stretch to say that every pharmaceutical company now studying cancer, HIV-AIDS, COVID-19, and a host of other diseases has made use of Jorgensen’s water models.

“The enduring success of Bill’s 1983 work lies in its rare combination of simplicity, transferability, and physical soundness, which one might see as a perfect embodiment of Occam’s razor in scientific modeling, or Leibniz’s principle of the best,” said Chris Chipot, research director at the Centre National de la Recherche Scientifique at the University of Lorraine, adjunct professor at the University of Illinois, Urbana-Champaign, senior editor of The Journal of Physical Chemistry.

Chipot is also a self-described “Jorgensen aficionado,” who has cited the 1983 study in nearly every one of his own papers for the past 30 years.

“The longevity of Bill’s models illustrates how fundamental, curiosity-driven research can have far-reaching and unanticipated impact,” he said. “It is a remarkable example of why supporting basic, methodologically sound research is vital — transformative tools often emerge from attempts to answer deceptively simple scientific questions. And we know that water is far from simple.”

Scott Miller, also a Sterling Professor of Chemistry in FAS and editor-in-chief of The Journal of Organic Chemistry, marvels at the ubiquitous nature of the work.

“The importance of water across chemistry, biology, and physics could never be overstated,” Miller said. “Bill Jorgensen’s seminal paper mapped out the fundamentals of simulating liquid water computationally, driving theory and experiment together, and impacting numerous fields.”

Of course, back in the early 1980s members of the scientific establishment were not tripping over themselves to hand Jorgensen the keys to the kingdom. Some older scientists had not fully embraced the ongoing computer revolution, which was still relatively new, as a practical way to explore chemical simulations and iterations of molecular interaction.

Jorgensen would have to prove that his approach was not only viable, but useful.

This is how he did it.

From chemistry sets to a computer revolution

Like many successful researchers, Jorgensen has been guided by scientific pursuits his entire life.

As a kid growing up in Port Washington, a hamlet on the western side of Long Island, and later in Sherman, Connecticut, he conducted scores of experiments with his trusty A.C. Gilbert chemistry set, tromping off to the local drug store regularly to replenish his supply of potassium nitrate. In high school, at Phillips Exeter Academy in New Hampshire, he took AP Chemistry and taught himself how to write computer code in BASIC.

He went on to graduate from Princeton in three years, learning the computer language FORTRAN in the basement of the Frick Lab and conducting work for his first co-authored study in the Journal of the American Chemical Society. Then it was on to Harvard for graduate school (where he worked with eventual Nobel winner EJ Corey), and to Purdue to begin his teaching and independent research career.

He soon came to focus his research on the need for better simulations of systems in solution.

“I realized very quickly that I wanted to study reactions in liquids and investigate the way molecules recognize each other in solution, which can lead eventually to drug design,” Jorgensen said. “In drug design you typically have an inhibitor, a small molecule that is binding to a disease-causing protein, which then disrupts the function of that protein.”

To do this work, he needed computing know-how and processing muscle. So, he taught himself statistical mechanics to go with his working knowledge of programming, and he got together funding for a research computer.

In the late 1970s, Purdue had two CDC 6400s (an early mainframe computer built by the Control Data Corporation) in its computer center. That meant two processors for 40,000 students and faculty, compared to today when the average laptop computer has eight processors. But times were changing.

“Fortunately, I was in the right place at the right time, because by the early 1980s computer resources became more available,” Jorgensen said. “You could have your own computer in your lab, if you could find the money to buy it.”

He and his research group purchased a Harris 80 — a tall cabinet computer that occupied its own room with an air conditioner and a printer. Much of the funding for it came from a 1978 grant to Jorgensen from the National Science Foundation (NSF).

“NSF was essential to the early work I did, including the water models,” Jorgensen said. “NSF funded basic science research that led to many of the technologies and therapeutics we take for granted today.”

Meanwhile, the older guard of scientists, who’d previously held computer modeling at arm’s length began to see the value of adapting to changing technology. “In the late 1970s you had people still trying to do paper-and-pencil theory work, but you also had the new people coming in with their computers,” Jorgensen said. “There was some difficulty in my being accepted by some of the theoretical chemists at the time, who were rather dismissive of what I was doing. I had to prove I could do something useful.”

Water world

Jorgensen spent months running the calculations for his water model masterworks.

With his Harris 80 and FORTRAN code, he was able to refine “force fields” for water — parameterized equations to describe the interaction energy of two water molecules from their atomic coordinates. Simulations for liquid water use the force field to evaluate all interactions between water molecules in a small sample of the liquid.

Researchers elsewhere were also developing force fields for proteins, but no one had created a good force field for water.

“The simulation community needed the water model because water is so essential to biomolecular systems,” Jorgensen said. “A key detail was that I wrote computer software to do simulations of liquids at constant pressure and temperature. That was unusual, but it let me compute the density of liquids — which is a critical property. Using the Harris 80, I was able to do iterative simulations to optimize the parameters in the model.”

It was slow going, even with his own computer. Jorgensen could only run one “job” at a time, submitting a slight change to the model and waiting until the next day to see the results — then changing a parameter here or there and waiting again. Fine tuning for density and for the heat of vaporization of water required many iterations.

He asked Michael Klein, then with the chemistry division of the National Research Council of Canada and now the Laura H. Carnell Professor of Science at Temple University, to collaborate on the research. Klein had been doing his own research on modeling liquids at the time and would carry out molecular dynamics work for the new models.

“Bill was more focused on chemistry applications, whereas I was more focused on the physics aspects, including the methodology of being able to do rigorous statistical thermodynamics on a computer,” Klein said. “So, Bill’s focus was more on comparing, refining, and optimizing water models, whereas I was more focused on ensuring that the computations themselves were rigorous. In that sense our collaboration brought together complementary skill sets.”

As part of the work, Jorgensen also ran experiments with existing, less-accurate water models, so that he could report side-by-side comparisons. To do this, he wrote the software for the simulations with the different models and did the iterative refinement for his models.

He called his models TIP3P and TIP4P, the “tip” meaning “transferable intermolecular potential.”

“We modeled water either as three points, where you have the oxygen and two hydrogen atoms, or as four points where we had an additional site,” he said. “A student and I developed an even better model with two additional sites — TIP5P — but that wasn’t until 2000.”

By then, Jorgensen had helped redefine the way medicines are discovered.

Elegant simplicity

As with all published, peer-reviewed research, Jorgensen’s water models went out into the world to sink or swim based solely on their own merit. He and his co-authors didn’t have to wait long for a verdict from the scientific community.

“The paper was timely and brought to conclusion a half century of research effort in the field of chemical physics,” Klein recalled. “It was immediately well-received.”

Indeed, the models proved to be extraordinarily effective in reproducing experimental data for liquid. Citations of the 1983 study in other scientists’ research began to pile up. By the end of the decade, a pair of key software programs used to calculate biomolecular dynamics — Assisted Model Building with Energy Refinement (AMBER) and Chemistry at Harvard Macromolecular Mechanics (CHARMM) — were using Jorgensen’s models.

“The TIP3P and TIP4P models capture the essential physics of liquid water with the fewest possible parameters, avoiding unnecessary complexity, while achieving excellent agreement with experiment,” Chipot said. “Their parsimony not only made them computationally efficient, but also broadly compatible with subsequent generations of biomolecular force fields. This elegant simplicity, grounded in physical realism, has allowed them to remain relevant for over four decades, even as more elaborate polarizable models have emerged.”

And Jorgensen continued pushing his research forward, including a seminal 1985 study on calculations of Gibbs free-energy changes (a thermodynamic quantity indicating the balance between forces in a reaction) in solution, the development of TIP5P, and leading work on computer-aided drug design.

The field of computer simulations for chemical and biological systems, meanwhile, has blossomed from tens of researchers in 1980 to thousands today. Applications for the field range from the design of materials for carbon capture and catalysis to the discovery of drugs for fighting viral infections, inflammation, pain, and cancer. Jorgensen’s influence is reflected in the fact that he has 18 published studies that have been cited more than 1,000 times.

He joined the Yale faculty in 1990 and has won numerous awards, including the American Chemical Society (ACS) Award for Computers in Chemical and Pharmaceutical Research, the ACS Hildebrand Award, the ISQBP Award in Computational Biology, the Sato International Award from the Pharmaceutical Society of Japan, the 2015 Tetrahedron Prize for Creativity in Bioorganic & Medicinal Chemistry, and the 2023 Arthur C. Cope Award for organic chemistry.

He was editor of the ACS Journal of Chemical Theory and Computation since its founding in 2005 through 2021, and he is a member of the American Academy of Arts and Sciences and the National Academy of Sciences.

At the outset of the COVID pandemic, Jorgensen and his Yale colleagues went to work applying decades of drug discovery research into rapidly developing potent inhibitors of the main SARS-CoV-2 protease (enzyme).

“Compared to 1980, when we could hardly model water, in 2020 we could, in a matter of months, design a potentially potent viral inhibitor,” he said. “It is unbelievable, really, to think back on where we were, and the struggles of doing a simple liquid simulation.”

If he ever needs a reminder, he need look no farther than the far wall of his Prospect Street office. On a high shelf, in a cardboard box, are the original computer printouts and handwritten notes from his 1983 study. Below them, on the floor, is a small computer more powerful than a fleet of Harris 80s.

“I have 80 jobs running on that computer cluster,” Jorgensen said. “You can’t even hear it, but it’s running simulations of molecules in liquid.”