Register Now: Advances in Spectroscopy for Pharma and Biopharmaceutical Analysis Symposium
News|Videos|November 17, 2026

How Multinomial Logistic Regression Reads Overlapping Blinking Statistics

In the below Eastern Analytical Symposium (EAS) interview, Kristin Wustholz of the College of William & Mary explains how machine learning applied to fluorescence blinking can identify dyes from cultural heritage objects without relying on SERS or spectral data.

Surface-enhanced Raman spectroscopy (SERS) is an ultrasensitive molecular fingerprinting technique, and Kristin Wustholz, a Margaret Hamilton Professor and Department Chair of Chemistry at William & Mary, has worked with it for many years.1,2

In the below interview clip from the Eastern Analytical Symposium (EAS), Wustholz describes the limitation that pushed her group toward a complementary approach: the difficulty of moving the analyte from a cultural heritage object directly onto the plasmonic substrate. Because the SERS signal depends so heavily on the distance between the analyte and the metal, poor contact means poor signal. Pretreatment methods can help, she notes, but they add complexity, and her group wanted a simpler route that avoids dye–nanoparticle interactions altogether.

That route turned out to be fluorescence blinking. Blinking has been discussed for decades, and although it is governed by a molecule’s underlying photophysics, it remains a stochastic, probabilistic process. Wustholz admits she initially doubted that intrinsic blinking could serve as a molecular barcode. Her thinking changed when students began identifying dye classes, such as rhodamines and anthraquinones, just by looking at blinking traces. Recent advances in data science and machine learning, she says, now make it possible to analyze blinking dynamics for patterns that would otherwise remain hidden.

The catch is that blinking statistics from different dyes form broad, overlapping distributions that look impossible to separate. Using multinomial logistic regression (MLR), the group evaluates those statistics simultaneously across multiple dimensions, essentially building a multidimensional regression. The model draws on more input predictors than a human analyst could weigh, and it can distinguish even structurally related molecules, such as rhodamine B and rhodamine 6G, without spectra or SERS.

The group did not start with MLR. Early work relied on a qualitative metric that Wustholz and her students developed through trial and error. Students with an interest in data science then pushed to try machine learning, and the team tested roughly 10 methods. Many performed well, but MLR offered an extra advantage: interpretable regression coefficients and probabilities that support more quantitative analysis of unknown samples. That interpretability is why the group has stayed with MLR.

References
  1. Wetzel, W.; Spectroscopy Staff. Previewing the Upcoming 2026 Eastern Analytical Symposium. Spectroscopy Online, 2026. https://www.spectroscopyonline.com/view/previewing-the-upcoming-2026-eastern-analytical-symposium (accessed November 9, 2026).
  2. College of William & Mary, Kristin L. Wustholz. College of William & Mary, 2026. https://www.wm.edu/as/chemistry/people/faculty/wustholz_k.php (accessed November 9, 2026).

Related to this article

Abstract concept of connecting molecules DNA network vector neurons | Image Credit: © bill2499 - stock.adobe.com
Garth Simpson of Purdue University outlines plans to survey more ordered molecular assemblies and to apply interface-specific chiral spectroscopy to enzyme reactors, solid-phase synthesis, spin-selective reactions, and, ultimately, a new analytical method.
Healthcare administrator reviewing an insurance policy schedule with a stethoscope on a clipboard, health insurance coverage, medical underwriting, prior authorization, and patient claims processing. | Image Credit: Joi_studio - stock.adobe.com
In a SciX interview, Marc Porter covered a lot of ground, from a student-driven pivot into immunoassays to a baseball analogy for Stokes and anti-Stokes scattering, and shares the open questions he is still chasing.