William & Mary: Chemistry partners with data science to develop a new investigative technique

Kristin Wustholz’s lab has developed a knack for melding chemistry with other disciplines, and the lab’s latest work incorporates a heavy data-science component.

The developers of single-molecule spectroscopy were awarded the 2014 Nobel Prize in Chemistry. The technique has been implemented in labs all over the world, and Wustholz believes that her refinement of the process will expand it even further.

“Multicolor single-molecule imaging is very widely applied, predominantly in biology — but also in material science,” Wustholz said. “Often, there’s a broad question like: What is the structure of this intracellular component?”

Wustholz is the Mansfield Associate Professor in the Department of Chemistry at William & Mary. Working with a group of student researchers, and supported by a grant from the National Science Foundation, she developed a new and improved approach to multicolor single-molecule imaging. Her group shared a proof-of-concept introduction, “Blinking-Based Multiplexing: A New Approach for Differentiating Spectrally Overlapped Emitters,” in The Journal of Physical Chemistry Letters.

Wustholz’s co-authors on the paper are a cadre of students, including Grace A. DeSalvo ’20, who is staying on as an M.S. student. Undergraduate co-authors are Grayson R. Hoy ’23, Isabelle M. Kogan ’24, John Z. Li ’20, Elise T. Palmer ’22, Emilio Luz-Ricca ’23 and Paul Scemama de Gialluly ’22.

Blinking-based multiplexing, or BBM, is an enhanced, data-rich variation of multicolor single-molecule imaging, which itself is based on the fluorescent emission of molecular components.

“Typically, if you have a sample that you want to image, you would stain it,” Wustholz said. “The stain adheres to different parts. Then you stick it on the microscope under a laser. But what you’re going to see on a microscope image is not a clear picture of the sample. All you will see are flashes of light as each of these molecules start emitting, then not emitting, then emitting again. A computer program puts all those flashes together to resolve the image.”
Isabelle Kogan and Grace DeSalvo are setting up the microscope for BBMIsabelle Kogan ’24 and Grace DeSalvo ’20 set up up the microscope for BBM.

Wustholz explained that problems with traditional multicolor single-molecule imaging stems from the limited number of fluorescent probes that can be used.

“If you’re trying to do a multicolor, super-resolved experiment, you’ll want to use three different colors,” she said. “So you have red, green and blue dyes to choose from. And as it turns out, if you’re trying to do biological samples, there aren’t a lot of dyes that play nice together.”

She went on to relate various workarounds that labs use to circumvent the mutual antipathy of dyes. Sometimes researchers use three different lasers. Other times, three different detectors. Or, she said, the labs engineer a sequential sampling regimen.

“Some institutions are able to purchase that gigantic instrument with 10 lasers and 10 detectors,” she said.

“The advantage here is we get rid of all that extra hardware stuff,” Wustholz explained. “We just have one laser and one detector.”

Wustholz ticked off other advantages, for instance BBM not only requires less instrumentation, but also opens up a new palette of dyes.

Multiplexing happens when the researchers look at the patterns flashing from the emitter. She said that the idea is to localize the molecule, which gives the resolution — then examine the pattern, to get the color of the molecule. The multiplexing translation process is a collaboration between humans and artificial intelligence, she added.

“My first instinct is to go with a human being; the human being knows the experiment,” she said, noting that human beings first recognized the blinking of the molecules had significance. “And so the human-being way is the taking all the data, coming up with statistics, trying to make sure that those statistics are separable. The AI way, the machine learning way, where you train the machine to differentiate the two. They both work.”

The lab brought on Luz-Ricca and Scemama de Gialluly, two non-chemistry majors from the university’s program in data science. “They really helped us with the machine learning,” Wustholz said. “That’s an exciting new part of the project that I’m excited to move forward.

“I have no experience in machine learning, and so I relied heavily on the students,” she added. “And, you know, when this went out for peer review, that part flew through with flying colors. So I really credit them with that. They were driven by their own curiosity and intuition. I think the combination of the foundational sciences like chemistry with data science is probably the future, and where we’re headed.”