Next Glass App Determines Tastes Based on Chemical Structure of Alcohol
The Next Glass app is available on smart phones and other devices around the US and Canada and aims to “predict the next glass of wine or beer that each person will enjoy.” According to Next App:
"Next Glass has developed the world’s first Genome Cellar, an extensive database that contains the chemical makeup - or 'DNA'- of tens of thousands of wines and beers. By looking at each bottle on a molecular level, Next Glass defines a unique taste profile for every bottle by analyzing thousands of chemical elements…
Once users download the Next Glass app, they are prompted to rate a few wines and beers that they have already tasted. Next Glass then compares their chemical taste preferences to the entire Genome Cellar to predict the next glass of wine or beer that each person will enjoy. Users just scan or enter the name of a wine or beer through the app and then receive a personal rating without ever having to take a single sip….”
Chemical Fingerprints for Thousands of Beers and Wines
So basically, magic happens in the lab and the app is stocked full of chemical “fingerprints” for thousands of beers and wines. The user downloads the app, answers a bunch of questions regarding what brands of wine or beer they like (or don’t like), then the app takes that information to create a unique “chemical profile” that will be indicative of the sorts of wines or beers the user is most likely to enjoy. Finally, the app takes that individuals profile and tries to match it against the chemical fingerprints it has on file for thousands upon thousands of beers and wines, and ultimately presents the user with the top beers or wines they would most likely enjoy.
According to Next Glass, they have been able to produce these individual flavor profiles and recommendations with 96% accuracy.
So how does the “magic” happen to produce the wine and beer chemical fingerprint library in Next Glass?
According to Trace Smith, COO of Next Glass, regarding the science behind the app:
“We spent considerable time and money (a year and six-figures) validating that we could translate chemistry and machine learning algorithms to taste. We had a contract lab run 100 bottles through every instrument one could possibly test wine with (more than a dozen). We got that data, layered it underneath a machine-learning layer, and held taste trials to determine efficacy of the data from each instrument. As it turns out, one instrument really works (and the rest help, but pale in comparison). That instrument is a high-resolution accurate mass spectrometer with a liquid chromatography pump (scientists call it an LC-MS). Ours is from Thermo Fisher and called an Orbitrap."
He goes on to say:
"Using mass spec to analyze wine is not a novel concept - scientific journals are full with articles on the subject. It's accepted science. The big wineries use mass specs for quality assurance (to make sure their products are consistent). The only real difference between their instruments and our Orbitrap is that ours is high-res, which gives you the precision and granularity you need to make recommendations (which is more than you need for quality assurance). It's like star-gazing at an observatory vs. using a telescope you can buy off Amazon. One is just much more powerful.
The characteristics won't be common to wine/beer drinkers. The outputs basically look like seismographs (they're call chromatograms). Those graphs translate to over 20,000 unique data points for each bottle we test.”
The Chromatograms
Whether looking at beer or wine, the procedures are still the same. I was sent an example of some beer chromatograms, in order to give you a visual of how Next Glass builds its library, and how it uses this library and your pre-determined preferences to pick other beers or wines that you would most likely enjoy.
This example is for three beers, Firestone Walker Pivo Pils, Pilsner Urquell, and Budwiser:
According to Smith, “The numbers above the peaks are the retention times, which help a viewer keep the peaks straight when the graphs are not overlayed. For example, if you look at the 2.80 retention time for Pilsner Urquell and Firestone Walker Pivo Pils, you'll see that Firestone Walker Pivo Pils has a lot more of the compound measured at the 2.80 retention time (The measure of the difference in relative abundance can be seen on the y-axis. These are all normalized, so they're being measured on the same scale (NL=2.90e7).
One thing to mention is these are very high-level views. With the software linked to our Orbitrap, we can 'zoom-in' for greater resolving power on any of these peaks to see sub-peaks that drive more subtle differences.”
To simplify even more, take a look at the X axis (parallel to ground) around the 2.8 mark. You’ll notice in all three graphs a bunch of peaks, some graphs having taller and greater number of peaks than others. According to PhD researchers for Next Glass, the compounds that show up around the 2.8 mark are related to “hoppiness” of the beer. Therefore, the hoppier the beer, the larger the peak. You can see in these graphs that the largest peaks around 2.8 are associated with the Firestone Walker Pivo Pils—this indicates that out of the three beers, the Pivo Pils is the “hoppiest”. Then, you’ll notice that the Pilsner Urquell has peaks not quite as high as the Pivo Pils at 2.8 but higher than the Budweiser peaks at 2.8, meaning that the Pilsner is less hoppy than the Pivo Pils, but more hoppy than the Budweiser. Finally, the Budweiser appears to have the smallest peaks around 2.8, indicating that this is the least hoppy of the three beers.
Smith notes, “Each of these chromatograms translates into the 20,000+ chemical attributes we collect for each bottle.” Meaning that for 20,000 sensory and chemical characteristics of each beer and wine, there is a graph just like the three shown above.
How does this all come together to predict the best wine for the user?
The hardest work of this app actually has to do with collecting all the individual chromatograms for the 20,000+ characteristics for EACH of the many thousands of wines and beers in the Next Glass database. After that, with just a little extra coding, the app pretty much runs itself.
First, you download the app.
Second, you “rate” as many beers and wines as you’d like in the Next Glass library.
Next, the app pulls the all the chromatograms for each of the beers and wines you preferred.
Then, the app runs some fancy mathematical equations to compare what beers and wines it KNOWS you like to the thousands of other beers and wines in the database, and pulls the ones that are closest in chemical profile to your known preferences and suggests “new” beers and wines for you to try and enjoy.
It’s a very simple program on the front side of things, but digging deeper into the science behind Next Glass, you can see that it’s not very simple at all, and is made up of thousands upon thousands of graphs and mathematical models to scientifically match the most appropriate beers and wines on the market with your own personal preferences.
Even if you’ve only had a couple of beers or wines from the Next Glass database, or if you’ve had the majority of them, it’s a great way for beginners to explore the world of beer and wine with more confidence, or for more seasoned drinkers to try something new without the worry of having to dump an entire bottle.