Industries / Food, farming and forestry / Food and beverage processing / Plant-based and alternative protein
Plant proteins
Plant protein work separates the protein in beans, seeds, grains and leaves from the starch, fibre and oil around it, as concentrated powders, purified powders called isolates, curds or drinks.
5
Milestones achieved, from c. 950 CE to 2023.
3
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
c. 950 CE
Bean curd as cheap mutton
Tofu is soy protein pulled out of ground, soaked beans by heating the milky liquid and curdling it, much as cheese is made from milk. The earliest surviving text to use the word, the Ch'ing I Lu written by T'ao Ku around 950, says people in one district called tofu "vice mayor's mutton", after a poor official who bought tofu every day because he could not afford meat.
Lessonthe story that the Han prince Liu An invented tofu has no early support; the Soyinfo Center finds no mention of tofu in the book attributed to him.
Credit
unnamed Chinese cooks; recorded by T'ao Ku
1939
Soy protein purified for food
In 1935 the Glidden Company's soya division in Chicago began making isolated soy protein, a powder of nearly pure protein, from solvent-extracted soybean flakes, for paper coatings. The first full-scale plant's product was "completely unsatisfactory" and the plant was destroyed by an explosion in October 1935. It was rebuilt in 1936 as a large pilot plant, and Percy Julian, hired that year as research director, led work on better quality. In 1939 Glidden introduced Albusoy, an enzyme-modified whipping agent that the Soyinfo Center calls America's first food-grade isolate.
Lessonthe extraction worked on paper years before a plant could make a usable product.
Credit
Glidden Company; Percy Julian and his team
1993
Enzymes make oats drinkable
Researchers at Lund University in Sweden filed a patent on a stable oat drink made by grinding oats in water and then letting two starch-cutting enzymes, beta-amylase and alpha-amylase, thin the slurry until it flowed like milk, without added stabilisers. The patent was granted in the United States in 1997 and passed to Oatly AB in 2009.
Credit
Lennart Lindahl, Inger Ahldén, Rickard Öste and Ingegerd Sjöholm
2013
Sorting pea flour by air
Wet extraction dissolves protein in water and recovers it with acid, heat and drying, which uses much energy and water. Researchers at Wageningen University milled yellow peas finely enough to free the small protein bodies from the larger starch grains, then separated them in a stream of air, a method called air classification. They reached concentrates of 51 to 55% protein, recovering up to 77% of the protein, with lower energy and water use. In 2020 a team in Italy and Estonia made meat analogues from dry-fractionated pea protein with about 55 g of protein per 100 g, noting few earlier practical uses.
Credit
Pascalle Pelgrom, Maarten Schutyser and Remko Boom (Wageningen University); Davide De Angelis and colleagues (University of Bari; Center of Food and Fermentation Technologies, Tallinn)
2023
A new protein plant fails
Merit Functional Foods began building a plant in Winnipeg, Canada, in autumn 2019 to make pea and canola protein ingredients with extraction methods developed by Burcon NutraScience; farms.com reported a projected cost of 150 million dollars. On 1 March 2023 its lenders, Farm Credit Canada and Export Development Canada, had a receiver appointed. The co-chief executive said the company "couldn't quite get to the level of cashflow needed to operate the business profitably, quick enough". Vegconomist reported debts of about CAD 95 million and a sale process that ran more than two years.
Lessona new extraction plant can run out of cash before it reaches steady sales, even with public lenders behind it.
Credit
Merit Functional Foods; Burcon NutraScience
Watch for
Leaf protein at 1,000 tonnes
Plantible Foods grows lemna, a small floating water plant also called duckweed, in enclosed ponds in Eldorado, Texas, and extracts rubisco, the most common protein in green leaves. It says new financing of USD 35 million will raise its yearly capacity for this protein fivefold, to more than 1,000 tonnes, with construction to finish in early 2027.
Achieved whensomeone outside the company reports the expanded site producing at that rate, or food makers report buying it in that volume.
Claimed by
Plantible Foods, August 2026
Rubisco from alfalfa
Leaft Foods says it extracts rubisco from lucerne (alfalfa) leaves in New Zealand at commercial scale without damaging the protein, which it calls a problem food scientists have worked on for more than a century. Its 30,000 ft² (about 2,800 m²) demonstration plant in Canterbury is reported to make about one tonne a week, and its claims on emissions, digestibility and function are its own.
Achieved whenan independent laboratory publishes purity and function measurements for the product, or a buyer reports steady supply.
Claimed by
Leaft Foods, August 2025
An open ingredient data garden
Food System Innovations, a US nonprofit, has launched the Data Garden, which it calls an open-access platform reporting solubility, foaming, gelation, emulsification, thermal behaviour and texture for plant proteins extracted and tested under the same conditions in its Berkeley lab. It starts with 10 ingredients, including pea, soy, hemp and fava, and is meant to grow into a public resource.
Achieved whenthe data can be downloaded freely and an outside group publishes work that uses them.
Claimed by
Food System Innovations, June 2026
Open
01
Lot-by-lot data on powders
A free, public table of how commercial plant protein powders behave, each measured by one set of methods and several lots: protein content, solubility at three pH values, gel strength, water holding, and emulsifying and foaming capacity, with the lot number of each sample. Add a simple model that predicts gel strength from the cheaper measurements.
The briefCredit · unclaimed02
Robot-run extraction trials
A small liquid-handling robot that runs hundreds of 2 mL protein extractions from one crop, for example faba beans or sunflower press cake, varying pH, salt, temperature and time, while an optimisation program picks each next batch of conditions to raise protein yield and solubility together.
The briefCredit · unclaimed
Open 01 · the brief
Lot-by-lot data on powders
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A free, public table of how commercial plant protein powders behave, each measured by one set of methods and several lots: protein content, solubility at three pH values, gel strength, water holding, and emulsifying and foaming capacity, with the lot number of each sample. Add a simple model that predicts gel strength from the cheaper measurements.
Why it is still open
The Good Food Institute says knowledge of how non-animal proteins behave is limited outside soy, pea and wheat, and that the databases NotCo and JUST built are proprietary. A 2024 Fraunhofer study of 61 commercial ingredients found that batches of the same product varied by up to 24% (relative standard deviation). The Data Garden, launched in June 2026, tests 10 proteins it extracts itself rather than commercial powders lot by lot.
Built with what exists
Standard food laboratory methods, following the test series proposed by Ma and colleagues at the University of Massachusetts in 2022; a texture analyser such as the Stable Micro Systems TA.XTplus for gels; a spectrophotometer and a commercial protein assay kit for solubility; scikit-learn or XGBoost for the model; Zenodo to publish the data under an open licence.
First step
Buy 10 pea and faba bean powders from five suppliers, measure solubility at pH 7 and water holding for each, and publish the table with the method.
Done looks like
At least 100 powders from at least 8 crops, 20 of them tested in three or more lots, all results and methods published under an open licence, and a model that predicts the gel strength of held-back powders with its error stated.
Open 02 · the brief
Robot-run extraction trials
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A small liquid-handling robot that runs hundreds of 2 mL protein extractions from one crop, for example faba beans or sunflower press cake, varying pH, salt, temperature and time, while an optimisation program picks each next batch of conditions to raise protein yield and solubility together.
Why it is still open
Extraction and drying conditions change both yield and behaviour, and a 2022 review found industrial isolates often less soluble than laboratory ones because commercial processing damages the protein. Searches of Europe PMC, GitHub and the web in November 2026 found self-driving laboratories for enzyme assays and plant compounds, but no published study or code that uses a robot and an optimiser to tune plant protein extraction.
Built with what exists
An Opentrons Flex liquid handler with deep-well plates, a plate centrifuge, a plate reader with a protein assay kit, and Meta's open Ax and BoTorch libraries for Bayesian optimisation, which chooses the next tests from the results so far. All chemicals are dilute acids, bases and salts at bench scale.
First step
Write and publish an Opentrons protocol for alkaline extraction and acid precipitation at 2 mL scale, and run 24 conditions on one faba bean flour.
Done looks like
At least 200 extractions run in one week and published with all conditions and results; the best conditions found reach a stated yield and purity, and repeat within 10% when scaled to a 1 kg bench extraction.