AI identifies hundreds of promising plant proteins for sustainable food ingredients

New research from NAPIC scientists at University of Leeds is using artificial intelligence and statistical physics to identify plant proteins that could potentially replace animal-derived emulsifiers.

Researchers from the University of Leeds’ School of Food Science and Nutrition have developed a new approach that can rapidly identify plant proteins with the potential to act as emulsifiers – a key functional ingredient used across food, cosmetics, pharmaceuticals and other products.

The model has identified nearly 800 plant proteins that could potentially act as emulsifiers, many of which had not previously been considered for this purpose.

The research could help address one of the major challenges in developing new sustainable food ingredients: how to identify promising proteins without having to test millions of possibilities individually in the laboratory.

Why do we need new emulsifiers?

Emulsifiers are essential for helping oil and water combine and remain stable. They are used in everyday products including sauces, ice cream and mayonnaise, as well as cosmetics, pharmaceutical products and industrial applications.

There is growing interest in developing natural and more sustainable alternatives to existing emulsifiers, including those derived from animal proteins such as milk proteins, including caseins and whey.

However, there are millions of potential plant proteins that could potentially have useful functional properties. Identifying the right ones through conventional laboratory testing can be expensive and time-consuming, relying heavily on trial and error.

The research team set out to find a faster and more reliable way of predicting which plant proteins could behave as effective emulsifiers.

Combining AI and statistical physics

The research was led by Dr Simha Sridharan and supervised by Professor Anwesha Sarkar, both from the University of Leeds’ School of Food Science and Nutrition and the Sarkar Lab. The team collaborated with Dr Rik Sarkar, a machine learning expert at the University of Edinburgh.

The researchers first used a simulation model based on statistical physics to understand how proteins interact with oil and water interfaces.

This interaction is important because, for a protein to work as an emulsifier, it needs to attach at the interface between oil and water and help stabilise the mixture.

The team then used machine learning to identify specific sections and characteristics of proteins that influence this behaviour.

By combining statistical physics and machine learning, the researchers were able to predict which plant proteins were most likely to demonstrate emulsification properties similar to animal proteins.

This computational approach means researchers can narrow down the huge number of potential plant proteins to those most worthy of further laboratory investigation – potentially reducing years of conventional trial-and-error testing.

Nearly 800 promising plant proteins identified

The model identified nearly 800 plant proteins with the potential to act as emulsifiers.

Importantly, many of these proteins had not previously been considered for this purpose.

The researchers then tested several commercially available proteins to see whether the experimental results matched the model's predictions.

The results were promising, with pea and potato proteins demonstrating effective emulsification properties, supporting the predictions made by the AI-driven approach.

Professor Anwesha Sarkar, NAPIC Co-Director at the University of Leeds, said:

The model identified nearly 800 plant proteins that could potentially act as emulsifiers, many of which had never previously been considered for this purpose.

The findings demonstrate how AI could help researchers identify promising new ingredients much faster than traditional experimental approaches.

What could this mean for the food industry?

The approach could be particularly valuable for companies developing plant-based and sustainable food products, providing a new way to identify functional ingredients from a much larger pool of potential protein sources.

Rather than relying solely on laboratory testing to work through potential proteins one at a time, computational approaches could help researchers identify the most promising options first.

This could accelerate the discovery and development of new ingredients while reducing the time and resources required during early-stage research.

The research also demonstrates the potential of bringing together different areas of expertise – including food science, protein chemistry, statistical physics and artificial intelligence – to address challenges in the transition towards more sustainable food systems.

For NAPIC, the work is an example of the interdisciplinary research needed to develop the next generation of alternative protein technologies and ingredients.