Using data science and artificial intelligence (AI) is the key to unlocking the latent potential that exists in the plant kingdom, says Oliver Zahn, astrophysicist and founder of plant-based dairy startup, Climax Foods.
There is no escaping the fact that animal agriculture is bad for the environment and that the food production system is going to sustain a growing global population. Global analysis has shown that while meat and dairy provide just 18% of calories and 37% of protein, they use 83% of farmland and produce 60% of agriculture’s emissions. In short, it is inefficient and unsustainable.
The food industry is all too aware of this, and is ploughing huge resources into the development of plant-based foods and technologies that have the potential to reduce reliance on animal agriculture. However, predictions of meat and dairy alternatives displacing demand for animal products haven’t quite materialised.
Oliver Zahn, founder of California startup Climax Foods, believes this is because so far, the alternatives are “pretty lacklustre”.
“Usually they don’t taste like animal products. They don’t have the same texture, they don’t have the same functional characteristics when you heat them, and they don’t have the same nutritional value,” he said, addressing delegates at the Future of Nutrition Summit at Fi Europe 2023.
Broadly speaking, two strategies are dominating advancement in this space: using trial and error approximation to develop products that closely resemble their animal-derived counterparts and using precision fermentation to grow cultured meat or dairy.
The problem with the former approach, according to Zahn, is that in trying to recreate the functionality of the animal product, companies usually end up with what he refers to as a “band-aid” product.
“They end up piling band-aids on top of one another. They might add flavours, then gums, and in the end, they still don’t manage to recreate the actual functionality,” he said.
He gave the example of casein, a dairy ingredient whose breakdown functionality is difficult to replicate.
“If you try to recreate the melt and stretch functionality of casein through a mix of ingredients that aren’t animal-derived protein, you will never end up with the same outcome. It may melt but it won’t melt in the same way,” he said.
Turning his focus onto cultured meat and dairy, Zahn questioned the assumption on which this approach is based, namely that the products created have to be biologically and chemically identical to their conventional equivalent.
“The notion that you have to recreate the animal product identically is an interesting route to explore, but ultimately the only thing that matters is that it behaves in the same way. It does not matter whether the protein from which ‘melt and stretch’ is derived has the sequence of casein in it. A human will not look under the microscope. No human cares. No human will ever care,” he said.
So how should the food industry be approaching the development of foods that can replace those produced by land-based agriculture?
According to Zahn, a better way is leveraging the largest resource of edible protein, lipids, and biodiversity on earth, namely the plant kingdom. This was the premise on which Zahn founded Climax Foods three years ago.
“The plant kingdom is so overwhelmingly large and complex that no company to date has dared to explore what we in data science call the ‘global optimum’ of what you can accomplish in terms of recreating any desirable food functionality,” he said. “Climax Foods is a data science company leveraging deep plant intelligence to unlock food products that are tastier, healthier and more sustainable.”
He said the key to unlocking this potential is to harness the intelligence of both humans and machines.
“There is a limit to what humans can imagine. There is also a limit to what machines and AI can imagine. The beauty of our model comes from the symbiosis of the two forms of intelligence,” he said.
He added: “And this is what we’re learning every month in our lab: how we can harness the intelligence of humans and machines simultaneously to speed up the product development process and answer the question of how we can replace animal products in a zero-compromise way more quickly.”
The starting point for this was to collect data on what makes cheese and other animal products behave the way they do in order to provide its AI-driven platform with a set of labels from which to learn.
“In supervised learning, you have a bunch of outcomes that are predicted by a bunch of inputs. So you need a comprehensive set of labels and features to be able to recreate these outcomes,” he explained.
With no existing data available, the company developed a comprehensive set of assays that characterised the human interaction with food – in other words, how humans perceive compounds that affect key functional attributes like flavour, texture, melt and stretch.
Essentially, this enabled it to create an AI-powered sensory panel, a far more cost effective and efficient solution for testing prototypes than using a human panel.
It has also provided the basis for rapidly screening thousands of ingredients and comparing the outcomes to its targets and animal equivalents. Machine learning enables it to converge faster to fewer failed experiments.
“We optimise focusing on functional efficacy, ingredient source, scalability and de-risked dependency on any one source ingredient,” explained Zahn.
Through this platform, the company is able to understand how proteins behave in response to physical and chemical modifications, and can optimise them to achieve certain characteristics.
For example, it can make a plant-derived milk matrix that it claims is the first protein matrix to stretch and melt like casein, and has developed a plant-based fat that combines the desirable properties of margarine and butter.
Zahn said that high throughput ingredient selection and proprietary protein modifications coupled with fat optimisation have enabled the company to achieve the specific melt and stretch properties of mozzarella cheese.
“We screened and modified thousands of ingredients in a data-driven way to find the ideal expression of functional properties. Then, with lipid databases and modelling, we created a plant fat mixture that emulated milk fat. We combined the fat and protein into a mozzarella prototype and used AI to optimise for different protein and fat combinations, delivering the best mozzarella fast,” he said.
As well as developing mozzarella, the company has used its platform to develop brie and blue cheeses.
One of the first questions it needed to answer when making cheese was whether it could use traditional dairy cultures even though it was moving away from lactose to plant-based sugars.
“We got quite excited about the prospect of developing new strains but actually found that the best way was to make a functional milk substrate that could use traditional dairy cheesemaking strains so that you can turn it into any cheese in exactly the same way,” said Zahn.
“Originally, we didn’t know whether the strains would prefer lactose over plant-based sugars. It turns out they don’t. They like plant sugars better, and sometimes they like them too much, so you need to slow them down a little,” he said.
This does have its benefits, though, as it means cheesemakers can speed up the ageing process - one of the biggest bottlenecks in cheese production.
Climax Foods is starting in plant-based cheeses, but its bigger mission is to develop a turnkey dehydrated replacement that replicates the functionality of animal milk, with the ultimate aim of comprehensively replacing dairy milk.
“If we can make a functional milk substrate that can be fermented, aged, and processed in the same way as milk, that would be very cool,” said Zahn.
