Crystals in a new light: UB team proposes rethinking crystal structure analysis

BUFFALO, N.Y. — University at Buffalo chemist Jason Benedict and his team spent years developing photoswitchable crystals.

Every crystal’s shape is a mirror of the internal arrangement of their molecules, but the molecules in photoswitchable crystals can expand, twist and change properties — from their color to their electronic conductivity — with a simple flash of light. This has made them highly sought-after for applications like pharmaceuticals and data servers. 

But scientists have very little control of the shape that crystals take, and after Benedict’s team painstakingly grew photoactive molecules known as dithienylethenes into 19 distinct crystal structures, just two responded to light.

“It was a real bummer,” says Benedict, PhD, professor in the Department of Chemistry, within the UB College of Arts and Sciences. “I can assure you, there’s not a lot of interest in non-photoactive crystals made from photoactive molecules.”

But, “after sitting back and licking our wounds a little bit,” as Benedict puts it, they began to ask some questions.

What makes some crystal structures conducive to photoactivity and others not? And why do so many photoactive molecules end up locked into the non-photoactive shapes?

To answer that, Benedict’s team has developed a new way of looking at crystals — not as isolated structures, but as data points in a much larger quantitative analysis. Their method, called the D–D analysis, can measure how photoactive molecules are arranged within tens of thousands of possible crystal structures, also known as crystal lattices. By turning those arrangements into numerical data, they can predict which structures are likely to be photoactive and, perhaps more importantly, why so many are not.

 “The Cambridge Structural Database, the world’s largest curated crystal structure database, has over 1 million crystals in it now. Let's move beyond thinking about a single crystal structure and begin to think about collections of crystal structures, and how their geometries might be related to photoactivity and other properties,” says Benedict, who is also the interim director of the UB Hauptman-Woodward Research Institute.

For their approach to really take hold, Benedict’s team says crystal research itself needs to change. This includes expanding the definition of what’s known as the crystal landscape to also include a crystal’s geometry, and encouraging scientists to publish studies on their crystal failures as much as their successes. 

“A lot of photoactive crystals are successfully grown, but it's largely serendipitous,” Benedict says. “We want to bring some intelligence to the process.”

Traditionally, the crystal landscape has referred to crystals’ energy. Scientists map a molecule’s possible crystal structures by their energy levels — low energy and stable on one side, high energy and unstable on the other. 

This is done extensively in drug research. Pharmaceutical companies often want to know the lowest energy crystal structure possible for their drug molecule.

Benedict’s team realized they could use their D–D analysis to instead build a crystal landscape based on geometry, comparing how a molecule arranges itself across many different crystals. 

Each crystal structure, once described numerically, becomes a data point in a map, revealing clusters of structures that have the same or very similar geometry. Scientists can then look for features that are common in one cluster but not in another.

“Once you understand those features, you can begin to see what conditions during the crystallization process might increase your odds of ending up with one particular structure over another,” Benedict says.

This would be a step toward crystal engineering — intentionally growing crystals with a specific molecular structure. 

“By and large, no one can do that yet. Generally speaking, nature tells you what kind of a crystal you're going to get,” Benedict says. “Hopefully our method is moving us all closer to crystal engineering.”

When most of Benedict’s crystals failed to respond to light, he compared his results to those in crystal databases and noticed something odd: an overwhelming — and potentially disproportionate — number of photoactive crystals.

“I applaud all the teams who have had success growing photoactive crystals structures, but I can't help but wonder if on their hard drives aren't a myriad of non-photoactive structures that just never made their way into the paper,” Benedict says.

Too many photoactive crystals and too few non-photoactive crystals can skew the data and throw off analysis. This is especially problematic for analyses done by artificial intelligence — machine learning tools are becoming more and more common in chemistry research. 

“We know that AI can give very biased results if the data that it’s trained on is biased. What you get out of it is only as good as what you put into it,” Benedict says. “Right now we have databases biased toward photoactive geometry, so you can imagine an algorithm concluding that a common feature of photoactive crystals is responsible for the photoactivity when that’s not actually the case.”

Benedict is calling on research teams to produce more studies on their unsuccessful crystals. His team did exactly that with their non-photoactive crystals, describing them in a study published in IUCrJ, a journal of the International Union of Crystallography.

“People aren't necessarily inclined to report their failures,” he says. “But the only way we're going to improve the odds of growing photoactive crystals is to understand why we get photoinactive crystals.”

The team has more recently used the D-D analysis to make a potentially troubling finding. 

They found that dithienylethenes — the photoactive molecules— don’t always crystallize into their low energy structures, as is typically expected. Instead, tight packing forces can actually push the molecules into high energy structures and hold them there, stabilizing an otherwise unstable shape.

Published in Crystal Growth & Design, a Journal of the American Chemical Society, the team’s finding upends the assumption that molecules tend to crystallize into their lowest-energy shapes because higher-energy shapes are too unstable.

“This would make crystal engineering — specifically, predicting crystal structures — all the more difficult,” Benedict says. “What’s the solution? That’s not clear. Crystals may still form on nature’s terms, but we hope our method will help us at least learn the rules.”