Summarize this content to 2000 words in 6 paragraphs
Humans’ noses can be adept at detecting delicate hints of apple or smoke in a glass of whiskey. But with machine learning, computers might also be Whiskey Masters.
By connecting the molecular makeup of 16 different whiskies with their odor profiles, a machine learning algorithm was able to identify the top five flavor notes in each whiskey, matching the flavors noted by a panel of human experts, scientists report December 19 in Communications Chemistry.
The bouquet of a drink is the product of dozens of gaseous molecules wafting up into the air. In the case of whiskey, there are more than 40 compounds that create odors ranging from vanilla to caramel to smokiness. The extensively trained human experts who can distinguish these flavor notes are called Whiskey Masters.
Scientists have hunted for laboratory methods to supplement that human expertise (SN: 10/27/22; SN: 10/31/19). Many tools, such as mass spectrometers, can identify the molecular makeup of whiskies. But getting from that makeup to the subtler impression of an array of odors has proven difficult, say data analyst Andreas Grasskamp and colleagues. Individual molecules can have different odors depending on the medium they’re in — air, water, oil — and different odors stand up against one another in complex ways.
Grasskamp — of the Fraunhofer Institute for Process Engineering and Packaging IVV in Freising, Germany — studies sensory perception processes. He and colleagues used a machine learning algorithm to test whether the molecular composition of whiskies could be used to predict their odor.
The test subjects were 16 previously analyzed whiskey samples, seven American and nine Scotch. To determine the top five flavor notes from each whiskey, the researchers combined two algorithms: one a statistical computer model that distinguishes samples based on the detected molecules, and the other a neural network trained to predict identifiable scents based on the detected molecules. Combined, the two algorithms then came up with the top five flavor notes in each whiskey.
The team compared these automated assessments with the tasting notes from 11 Whiskey Masters, each of whom had identified the top five odors they detected in each whiskey out of 17 preselected attributes. Everyone didn’t come up with the same top five for each whiskey, so the researchers determined an aggregate top five odors per whiskey.
The top five flavor notes identified by the algorithm consistently matched the aggregate top five identified by the humans — suggesting, perhaps, greater accuracy in identifying the strongest aromas.
But however good a computer Whiskey Master might be at identifying flavors, it still can’t tell you how much you’ll enjoy them.
