INFORM February 2025 Volume 36 (2)

10 • inform February 2025, Vol. 36 (2)

QUANTIFYING TASTE At IFF’s Physical Food Science group, Flemming Møller is one of the scientists generating this type of data, or “putting numbers to food,” as he said. The numbers come from measurements using a variety of techniques—microscopy, rheology, X-ray dif fraction, and others. Even when the techniques themselves are not particularly new, improvements have led to an explosion in data. Møller points to confocal microscopes, which his lab uses to inspect materials at the particle level. A new version of the microscope allows for imaging with a thousand wavelengths of laser light; the previous one at his lab had three. “Today the instruments are spitting out so many num bers, we generate more and more data,” he said. “Combining the different instrumental results and linking them to product quality is highly dimensional. We use a lot of machine learning for that.” The finer-grained data makes it easier to distinguish among different crystal forms of fat, for example, or to see how a protein stabilizes a fat in an emulsion. These factors can affect not only product quality but the best processing and storage conditions. “Machine learning enhances our ability to understand how interactions between composition, processing conditions, taste, and stability affect the final product,” Bhattacharya said. “By analyzing large datasets, machine learning algorithms can identify patterns and relationships that might not be evident

“Some people come to me and say, ‘Can I predict the solid fat content that would result from the composition of my triglycerides?’” said Alejandro Marangoni, a professor and Tier I Canada Research Chair in Food, Health and Aging at the University of Guelph in Ontario, Canada. Marangoni is editor-in-chief of the AOCS Lipid Library, and his research lab developed an update for the library’s Triglyceride Property Calculator, a “melting property generator” that will predict enthalpy and temperature for different triglyceride types. “We have theory and we have data,” he said. “The inter section between the two allows you to make predictions. That is a perfect example of the capability of data science.” Such analyses go beyond simply evaluating materials or their combinations based on generic melting profiles. In the case of replacing trans fats, food producers have used tech niques such as interesterification, where fatty acids on the glycerol backbone of the triglyceride are rearranged, or fats are blended or fully hydrogenated. This creates a complex dia gram of potential fat combinations and interactions. “People from industry would love to know the solid fat content of interesterified fats at different temperatures without having to first synthesize and then measure them,” Marangoni said. “You can actually use data science for opti mization, and that would save a ton of time and money.” He noted that mixing individual triglycerides experimentally can cost hundreds of thousands of dollars based on the cost of materials alone.