Climax Foods uses AI to design plant-based cheese and is expected to officially sell brie and feta

When Oliver Zahn, CEO of Climax Foods, served me a plate of vegan brie, feta and blue cheese, my expectations weren't high.

Most vegan cheeses are only at the edible stage, and their taste and texture are still far from real cheeses. But the brie I tasted today had a smooth, full-bodied, velvety texture, and it was delicious.

If no one had told me, I would have mistakenly thought it was made from milk, but in reality it was made entirely from plants. Zahn told me that it wouldn't have been possible without machine learning.

Screenshot of Climax Foods' official website (source: Climax Foods)

Climax Foods is one of several startups that use AI to design plant-based foods, with others including Shiru in the United States and NotCo in Chile.

These companies train algorithms on ingredient datasets that contain desirable characteristics such as flavor, aroma, or stretchability. They then used artificial intelligence to comb through vast amounts of data to develop new combinations of ingredients that tasted similar.

Shiru's CEO, Jasmin Hume, wrote a PhD thesis on protein engineering.

"The traditional ingredient discovery process can take years and tens of millions of dollars, and the results are only slightly better than the previous generation," she said. [Now] we can start from scratch, pick out the best performing proteins, and prototype and test them in about three months. ”

Not everyone in the industry is bullish on AI-driven ingredient discovery technology. Food consultant Jonathan McIntyre, who previously led PepsiCo's R&D team for beverages and snacks, believes the technology is "obviously" overblown in his field.

"AI can only do as good as the data you provide," he said. He added that given that food companies are very protective of formula and proprietary information, researchers may not necessarily have enough data to produce productive results.

McIntyre tells a story about PepsiCo's efforts to make better soda using IBM's artificial intelligence, Watson. "It gives the worst-tasting thing ever," he said. ”

Climax Foods circumvented the data shortage by creating its own training set to fundamentally reverse engineer why cheese tastes so good.

Zahn, a former head of data science at Google Ads, said: "When we started, there was very little data on why animal products tasted the way they did, cheddar, blue, brie and mozzarella, and we had no commercial incentive to figure out what they tasted like. ”

At Climax's food science lab, Zahn showcased some of the instruments his team used to build a "treasure trove of data." There is a machine that uses ion chromatography to show the precise balance of different acids after a bacterial strain breaks down lactose.

A mass spectrometer acts like an "electronic nose" that reveals which volatile compounds produce our olfactory response to food.

A device called a rheometer tracks how cheese reacts to physical deformation, and our preference for cheese depends in part on the texture when chewed.

This cheese data creates what is known as a "baseline of performance targets" where AI can try to reach targets using different combinations of botanical ingredients.

Over the past four years, Climax's food scientists have analysed which plants might be good performing substitutes based on clues given by the tool and created more than 5,000 cheese prototypes.

Using the same laboratory instruments used to measure animal cheese, the Climax team performed an analysis that included about 50 different texture and flavor determinations, generating millions of data points in the process.

AI tools are trained on this data, and the algorithm comes up with combinations that might perform better, and then the team keeps trying new combinations and iterating.

"You change all the input parameters, measure the output of the outcome, and then try to narrow the difference between the outcome and the animal cheese to the smallest possible extent," Zahn said. ”

He estimates that Climax has analysed about 100,000 plant component combinations, including small-scale "miniature prototypes."

Zahn says it would take thousands of years to rely on human tasting and subtle tweaking of the ingredient combinations in so many prototypes.

But just over 3 years have passed since 2020, when he and his team have formulated and sold their first piece of cheese (their product went on sale in April 2023).

The plant-based ingredients of this vegan blue cheese are hardly special. The top four ingredients are pumpkin seeds, coconut oil, lima beans, and hemp seed protein powder.

However, Michelin-starred chef Dominique Crenn describes it as "soft, buttery, and surprisingly rich, beyond my own imagination of vegan cheese".

Zahn's company has reached an agreement with Bel Group, a well-known food manufacturer, to issue a license for the company's products. Zahn has also signed an agreement with another large producer whose name has not yet been disclosed.

Zahn is currently in discussions with venture capital firms to raise money and hopes to start selling brie and feta cheese later in 2024.

Unlike IBM Watson, which tries to formulate better recipes for soda, Climax's algorithm combines different ingredients in an alchemy-like format.

"The interaction of one ingredient with another can elicit an unexpected taste or texture," Zahn says. It's not just the sum of the two ingredients, it's something completely different. ”

One reason for developing dairy cheese alternatives is its environmental cost. Cheese has a higher carbon footprint by weight than chicken and pork, and humans consume around 22 million tons of cheese each year.

For Zahn, what he is looking for is not to satisfy consumers with a rubbery, poor-tasting alternative, but to offer a plant-based version that tastes just as good, if not better, and is less expensive to make.

Support: Ren

Typesetting: Liu Yakun