Deep Dive: Help is at hand

Plant-based, cell-cultivated and fermentation-enabled foods are receiving a high-tech boost in the digital age. Here, David W. Smith dives into how AI and machine learning are revolutionizing smart manufacturing in the alternative protein industry, making sustainable food production more efficient and delicious

The alternative protein sector is full of spirited people leading startups they hope will change the world. Passionate about sustainability, often vegan or vegetarian, they want to use advanced, or ‘smart’ technologies to revolutionize the food industry. One of the most fervent voices belongs to Yossi Quint, Founder & CEO of US company, Ark Biotech, which is developing biotech infrastructure to sell to cultivated meat companies wanting to scale up. Quint describes cultivated meat as the “most important landmark invention of our time” and believes bioreactors are critical to reaching the industry’s potential by bringing down exorbitant manufacturing costs.

“The future of meat is cultivated and will dramatically change the world on the most monumental scale,” he says. “Around half the world’s land is used for agriculture, according to the UN, and two thirds is grazing land. If you replaced that with bioreactors creating cultivated meat, or fermenters, you could return all that land back to the natural environment. It would dramatically reduce human suffering and provide affordable, nutritious food.”

Replacing just 1% of the world’s meat consumption would necessitate a 100-fold increase in bioreactor capacity, however. What’s more, the bioreactors would have to be designed to make cultivated meat, as opposed to the majority in use today that are designed for pharma. Ark Biotech, Quint says, has designed a range of bioreactors capable of scaling to millions of liters, whereas the biggest bioreactor today is 30,000 liters. At scale, Ark’s bioreactors will be 85% cheaper than the biggest bioreactors today.

“We see scale as the biggest single lever to reduce costs. In terms of operational complexity, it’s simple to operate fewer bioreactors because you use the same sensors whether it’s 10 or 10,000 liters. Samples need to be checked and it requires fewer people when you scale up,” Quint continues. “Larger systems offer economies of scale. We’ve also put a lot of effort into ensuring the process is sterile. Contamination is a common problem in biopharma as the cells are so sensitive.”

Ark has designed an automated operating system for bioproduction, which relies on real-time data from sensors and a digital twin to take autonomous decisions. “We’ve also developed computational models that help companies to scale between bioreactors. Right now, they could spend years going from small to large scale, then onto production scale,” says Quint. “This is partly because traditional bioreactors use impellers and impeller-based systems are hard to scale as the fluid dynamics are so complex. There’s a lot of trial and error. With our system, the fluid dynamics are much more predictable across scales.”

Ark primarily uses AI for two purposes. One is in process development – to determine the optimal strategies for running the bioreactors by analyzing the sensor data. The second is to monitor production in real-time.

Traditional bioreactors use impellers and impeller-based systems are hard to scale as the fluid dynamics are so complex

‍Design assistance

For other companies, AI is used to develop the ingredients for the feed-stocks going into cultivated proteins and fats. AI speeds up the fantastically complex processes and reduces costs. Multus, a UK company, is using AI to design feed-stocks for cells. “Right now, the costs of development are way too high,” suggests Dr Charlie Taylor, Multus’ Head of Business Development. “Scientists have grown animal cells and culture for decades, but in small quantities for regenerative medicine and, recently, in the biomedical arena for cancer treatments. But it’s been really expensive, although they’ve only needed small quantities. When it comes to food, we need a lot and the feed-stock is the biggest cost driver. It costs £200-£300 (US$252-US$379) a liter and you need 50 liters to produce a kilogram of cells to turn into food. The cost has to be shifted several decimal places to make the industry viable.”

Making feed-stock is also time-consuming as it contains between 50 and 100 ingredients. There are essentially two buckets – the first includes basic ingredients of around 20 amino acids, vitamins, minerals, fat and salt. The second bucket contains biologically active molecules. They are proteins that signal to cells it’s fine to keep replicating. They stabilize the environment around the cells. But the feed-stock needs to be optimized for each application, which is fiendishly hard using old-fashioned methods.

“To give you an idea of the complexity, imagine being given a kilogram of each ingredient to make a Yorkshire pudding – egg, milk, flour, salt, oil – you have to figure out the relative proportions and the right oven temperature,” Taylor continues. “You’ve only got a few variables, but you’d still need a lot of experiments. If you want to be fancy, throw in some rosemary and pepper. Each time you add a variable, it increases the number of experiments. With 50 to 100 it becomes an astronomical number, which is called an optimization challenge. But the vast majority of the 150 or so cultivated meat companies are using the traditional method called ‘Design of Experiments’, which has been the way for 100 years. That is like ‘taking a knife to a gunfight’. The scale of the problem dwarfs the potency of the method.”

With machine learning, computers carry out the heavy lifting. The average company has the bandwidth to do between 50 and 150 experiments in parallel, but AI allows 3,000 parallel experiments. “We do fewer experiments because machine learning helps decide which ones to do, and we don’t have to do as many sequentially as they can be done in parallel. It’s far cheaper, faster, and more scalable,” Taylor says. “With conventional methods they grow cells for two or three days under a microscope, then count them, giving one data point. We take a photograph every three hours and send it to a machine-learning application to count the cells. So, we’re getting a data point every three hours.”

Measured and made

Hoxton Farms, a London, UK-based company founded in 2020, is using AI to develop cultivated fat, but has a very radically different perspective to Ark on the development of bioreactors. Rather than build much larger bioreactors, Hoxton Farm’s bioengineers have manufactured smaller models that the company’s Ed Steele, Co-founder, says limit contamination and the loss of batches, saving millions of dollars in investment costs.

“A lot of companies have struggled to scale up bioreactors effectively,” says Steele. “You can start with 100 liters, then go to 1,000 liters, then 10,000, but you have to redesign the biology each time as everything changes. Instead, we build small bioreactors and place lots of them alongside each other. The challenge for us is to make a very cheap, small bioreactor,” he adds. “We need to perform a lot of operations on the reactors, such as adding new media, or taking samples, and it’s easier to automate that process if you have a lot of smaller reactors rather than fewer larger ones.”

Hoxton Farms has built a new 14,000ft² facility in the Finsbury neighborhood of London, which was funded by a US$22 million Series A funding round in October 2022. Two thirds is devoted to cell culture labs, a food development kitchen and hardware workshop. The initial capacity is to produce a ton of cultivated fat per year, with potential to scale up to 10 tons. The company’s initial focus is on creating a cultivated alternative for pork belly, but later it could develop chicken and fish variants. Steele says a few drops of cultivated fat can significantly enhance the taste and texture of plant-based products.

Hoxton Farms’ engineers takes continuous measurements of how the cells are behaving to adapt the parameters accordingly. “We’ve built a monitoring system to supply answers to questions such as, what is the temperature and the pH? How many cells are in our culture and how are they growing? Are they still viable? When will they start turning into fat, and how big and juicy do they get? What’s the fatty acid profile? And so on,” he says. “Then we can change parameters like the temperature of our reactors, how quickly we’re mixing, what nutrients are in the media, or if we need to add supplements.”

Search to the rescue

AI is also a critical tool in the far more industrialized sector of plant-based meats. The challenge there is to find ingredients in nature that mimic the properties of animal fat. Last year, the US biotech startup, Shiru, announced the commercial launch of OleoPro, a plant-based protein ingredient developed using AI. It says OleoPro holds its shape at room temperature, browns when cooked, and delivers a juicy, fatty mouth-feel in plant-based meat applications. There is up to a 90% reduction in saturated fat, too.

Without AI, the new ingredient would have been in development for many more years. Alternative proteins are derived from plants, algae, fungi and potentially produced via fermentation. But there are billions of distinct proteins, and finding the right solution is “like looking for a needle in a haystack”, according to Jasmin Hume, Founder & CEO.

“Thanks to machine learning and the availability of large datasets, we are in the middle of a revolution in how biological structure, function and interactions are inferred,” she says. “Shiru uses a combination of protein language models (PLMs) and other machine learning tools to rapidly sift through hundreds of millions of proteins and identify the best functional proteins for any given application. This is game-changing. Rather than spend years and millions of dollars on empirical tests, we arrive at the best solution in a fraction of the time and cost.”

Shiru debuted OleoPro in March 2023 in a plant-based chicken karaage, developed by partner Griffith Foods’ corporate venture arm, Nourish Ventures. The acceleration of the developmental process, thanks to machine learning, will enable the alternative protein sector to reach consumers more affordably and faster. “The sector needs to shorten the time and capital investment it takes for discovery, development and manufacturing of alternative proteins,” continues Hume. “Shiru is tackling the discovery element and derisking the economics to some extent. The industry needs to invest in automation, data-driven optimization, and management of complex flows.”

Rather than spend years and millions of dollars on empirical tests, we arrive at the best solution in a fraction
of the time and cost

Rockwell Automation is also using AI to evaluate every element of the production process in both plant-based meat production and cultivated meat production. “In plant-based meat production, this includes the protein, fat, fiber, and moisture content of various ingredients, alongside processing parameters like grinding temperature and fermentation conditions. The focus then shifts to product quality metrics such as texture, color, and nutritional profile,” says Todd Gilliam, Americas Industry Manager for CPG, Rockwell Automation. “For cultivated meat, it requires close attention to cell health (viability, growth rate), bioreactor conditions such as temperature, pH, oxygen, and ensuring proper muscle fiber formation in the final product.”

For quality control, AI can be trained with sensory data to evaluate taste, texture, and appearance of alternative protein products. Artificial intelligence and machine learning are revolutionizing manufacturing, Gilliam feels, by helping identify “golden batches” with perfect production output. “This allows for consistent quality across production batches and helps identify areas for improvement. By leveraging artificial intelligence and machine learning, alternative protein producers can streamline development, replicate the golden batch conditions, optimize production processes, and ensure consistent high-quality products for consumers.”

Efficiency drive

If we take a simpler and broader definition of ‘smart manufacturing’ as using methods that are more efficient and cost-effective, Rebellyous Foods, in the USA, is an impressive example. The company was founded in 2017, in Seattle, by Boeing engineer, Christie Lagally. Her aim was to increase production capacity and lower costs in the plant-based meat industry. Lagally designed the ‘Mock 2’ automated and chilled manufacturing process system that enables the production of 2,500 lbs of plant-based chicken an hour with potential to reach 5,000 lbs/hr, or more, on a single processing line in a continuous rather than a batch process. Lagally says this is on a par with conventional chicken processing. Mock 2 is at pilot scale, but will be installed into a facility in the near future.

“About 95% of all plant-based food is made using what we call the ‘mix and form’ method, where we take texturized vegetable protein, mix it with water, oil and starches into a substrate that looks like the ground meat product. Then we form it into hamburgers, or chicken nuggets, or hot dogs. The problem is almost all of it is done in meat-processing facilities that use the wrong tools for the job, especially as you scale up,” believes Lagally.

First, a big tenderizing tool designed to chop up meat is used to mix plant-based meat and hydrate the textured protein, she continues. Second, it’s necessary to emulsify oil, water and starch as plant-based foods don’t have cholesterol and have low saturated fat. Then, the batch has to be separated into different containers.

“But the ingredients are very thick and hard to mix, and it can get very warm, which decreases flavor.

“Another issue is the material sticks to the walls of the mixer and containers, making it hard to extract, whereas animal meat just slides down. Then, it is pressed into various shapes and breaded. It’s a laborious process and even when you scale up, it still costs three times as much as conventional meat.”

With Mock 2, the texturized protein is hydrated continuously, reducing labor costs dramatically. “Nobody has to scrape things down. All the material is hydrated, or emulsified, and mixed automatically so you just get dough,” Lagally explains. “It’s about industrializing the production of plant-based meat, which is our version of ‘smart manufacturing’. Essentially, we’ve designed the right tools for the job. We want to do what the meat industry did in 1961 when the automatic eviscerator was invented. No number of humans could handle every one of the carcasses in a modern chicken processing line,” she says.

In the USA, sales of plant-based meat fell by 12.2% year-on-year to US$76.7 million in September, 2023, according to Circana. Large companies such as Beyond, Impossible and Meati have laid off staff. Rebellyous has a different business model though: it supplies more than 200 school districts under the USDA National School Lunch Program with breaded soy-based nuggets, tenders, and patties. This market has been growing fast as more schools demand plant-based options. Rebellyous estimates a 90% reduction in production workforce costs and an 80% reduction in energy costs.

“Meat facilities are always chilled to -40°F as the slaughter of animals introduces a lot of bacteria on the surface of the meat,” Lagally continues. “But plant-based meat dough is highly viscous and insulating. It doesn’t have the same thermal properties so all we need to do is chill the equipment instead of the whole facility.”

We want to do what the meat industry did in 1961 when the automatic eviscerator was invented

Another benefit is the process is kinder, physically, on the human body than meat processing. “You won’t get repetitive strain injuries as there is no rotating blade equipment. Workplace chilling is another bad worker injury issue in meat processing. US Bureau of Labor Statistics show workers in animal slaughter and production face higher rates of injury than coal miners or construction workers.”

Leveling the playing field

Elliot Swartz, Principal Scientist for Cultivated Meat at the Good Food Institute, believes smart manufacturing is the best way for alternative proteins to compete on a bigger scale. And he says the sustainability benefits can be huge due to the reduction in land use. “Agricultural sprawl is using so much land and contributing so much carbon through deforestation that if you’re able to have more alternative protein diets, you would shrink the land footprint considerably,” he says.  

But alternative protein manufacturing will remain energy intensive and renewable energy should be part of every company’s equations, especially the use of onsite solar. Also important, he adds, is an increase in ‘circularity’ within the bio economy. “For example, a lot of raw material used to grow soybeans can be used to make plant-based meat. You can break it down into amino acids to feed to cultured animal cells. It’s the same with sugars, like from corn or sugar beet. You can use glucose to feed microbes or cells,” Swartz notes. “A good idea is to co-locate facilities, such as fermentation technology, where raw materials are being processed.”

The urgency of finding ways to manufacture cultivated meat, in particular, is linked to the global growth in meat consumption as GDP increases. The countries that are becoming richer will consume more meat. According to a 2023 study by CE Delft, replacing farming animals with cultivating meat from cells could cut the climate impact of meat by up to 92%, reduce air pollution by up to 94%, and use up to 90% less land.

“A lot of countries in Africa and Southeast Asia – where most of the population growth is expected over the next few decades – will prompt a rise in meat consumption globally between 50% and 100%,” Swartz says. “Cultivated protein has to be a major part of resolving this issue as we cannot keep expanding the agriculture sector.”

‍

‍