Researchers at the University of Leeds have used artificial intelligence and statistical physics to identify nearly 800 plant proteins capable of acting as emulsifiers — a discovery that could reshape how the industry sources alternatives to animal-derived ingredients like caseins and whey.

Solving a trial-and-error problem

Emulsifiers are essential for helping oil and water combine and remain stable, underpinning everyday products from sauces and ice cream to mayonnaise, as well as cosmetics and pharmaceuticals. While interest in natural, sustainable alternatives to animal-derived emulsifiers has grown, the sector has faced a fundamental bottleneck: millions of plant proteins could theoretically work, but identifying the right ones through conventional lab testing is expensive, slow, and reliant on trial and error.

A team from the University of Leeds’ School of Food Science and Nutrition, led by Dr Simha Sridharan and supervised by Professor Anwesha Sarkar, set out to build a faster, more reliable prediction method.

Physics meets machine learning

The approach combined two disciplines. First, the researchers used a simulation model grounded in statistical physics to understand how proteins interact with oil and water interfaces — the mechanism by which a protein stabilises an emulsion. They then applied machine learning to pinpoint the specific protein sections and characteristics that drive this behaviour.

Together, the two methods allowed the team to predict which plant proteins were most likely to emulsify in ways similar to animal proteins, screening out unpromising candidates before any lab work began. The project also drew on machine learning expertise from Dr Rik Sarkar at the University of Edinburgh, alongside NAPIC researchers Professor Nik Watson, Maryam Afzali and Thomas Hazlehurst.

Nearly 800 candidates, many previously overlooked

The model flagged close to 800 plant proteins with emulsifying potential — a significant share of which had never previously been considered for the role. To validate the predictions, the researchers tested several commercially available proteins against the model’s output. Pea and potato proteins both demonstrated effective emulsification properties, matching the AI-driven forecasts.

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.”

Next generation

For formulators working on plant-based and sustainable products, the value here isn’t just the specific proteins identified — it’s the method itself. A computational filter that narrows millions of candidates down to a testable shortlist could meaningfully cut the time and cost of early-stage ingredient development, reducing dependence on exhaustive lab screening.

It’s also a signal of how interdisciplinary research — spanning food science, protein chemistry, statistical physics and AI — is starting to shape the next generation of alternative protein ingredients. 

The research, “Data-driven pipeline enables discovery of plant protein surfactants,” is published in Communications Chemistry.