PROTSENS Rethinking Alternative PROTein Extraction: Decoding SENsory-Protein Extraction Relationships
Lead Research Organisation:
University of Leeds
Department Name: School of Food Science and Nutrition
Abstract
The global food system emits about 18% Green House Gas (GHGs), due to its reliance on animal-sourced ingredients such as proteins.
Therefore, to reduce the environmental impact of the food system and improve sustainability, plant, insect, and microalgae-based
protein sources are being investigated. However, these proteins suffer from poor sensory perception of bitterness and astringency,
leading to poor adoption by society. Previous research projects have failed to provide mechanistic insight and strategies to minimize
these negative sensory perceptions at an ingredient level. Therefore, the question remains, whether these proteins are inherently
bitter and astringent or does the protein extraction process leads to protein modifications that cause these sensory perceptions.
PROTSENS aims to understand and minimize these negative perceptions using green extraction methods that preserve protein
structure. Protein extracts resulting from these novel processes will be thoroughly characterized. The mechanism of astringency and
bitterness will be studied from macro to nanoscale using physical methods, while bitterness will be investigated using sensory cells in
vitro in a simulated oral environment. The in vitro, ex vivo and physical tools will serve as a novel toolbox to investigate astringency
and bitterness perception. The link between protein properties, extraction methods, and their sensory perception will be captured in
a Machine Learning (ML) model. The model will provide a guide on protein extraction routes, based on protein source to preserve
protein structure and improve sensory perception. Overall, PROTSENS, through its novel experimental toolkit and ML model,
improves the sensory perception of novel proteins and adoption. PROTSENS will open new horizons in European academic research
on food proteins. The success of PROTSENS will help drive the adoption of novel proteins, decrease GHGs in the food system, and help
mitigate climate change.
Therefore, to reduce the environmental impact of the food system and improve sustainability, plant, insect, and microalgae-based
protein sources are being investigated. However, these proteins suffer from poor sensory perception of bitterness and astringency,
leading to poor adoption by society. Previous research projects have failed to provide mechanistic insight and strategies to minimize
these negative sensory perceptions at an ingredient level. Therefore, the question remains, whether these proteins are inherently
bitter and astringent or does the protein extraction process leads to protein modifications that cause these sensory perceptions.
PROTSENS aims to understand and minimize these negative perceptions using green extraction methods that preserve protein
structure. Protein extracts resulting from these novel processes will be thoroughly characterized. The mechanism of astringency and
bitterness will be studied from macro to nanoscale using physical methods, while bitterness will be investigated using sensory cells in
vitro in a simulated oral environment. The in vitro, ex vivo and physical tools will serve as a novel toolbox to investigate astringency
and bitterness perception. The link between protein properties, extraction methods, and their sensory perception will be captured in
a Machine Learning (ML) model. The model will provide a guide on protein extraction routes, based on protein source to preserve
protein structure and improve sensory perception. Overall, PROTSENS, through its novel experimental toolkit and ML model,
improves the sensory perception of novel proteins and adoption. PROTSENS will open new horizons in European academic research
on food proteins. The success of PROTSENS will help drive the adoption of novel proteins, decrease GHGs in the food system, and help
mitigate climate change.
People |
ORCID iD |
Publications
Sridharan S
(2026)
Data-driven pipeline enables discovery of plant protein surfactants
Sridharan S
(2025)
Artificial intelligence in colloid and interface science: Current research, challenges and future directions
in Current Opinion in Colloid & Interface Science
| Description | Protein adsorption to hydrophobic surfaces is not related to number of hydrophobic amino acids (AA) at the protein level- the resolution needs to be fine-tuned to smaller block level. i.e: The presence of hydrophobic amino acids only affects the local adsorption rather than at whole protein-level. Adsorption at block level is a multi-variate phenomenon impacted by the relative abundance of hydrophobic, charged and polar AA. Adsorption can be predicted by segregating proteins into smaller blocks and extracting their amino acid compositional features Protein Language models such as ESM and ProtTrans often fail to predict adsorption and cluster proteins based on their adsorption properties. |
| Exploitation Route | We have deve;oped a data-driven pipeline, that can be used for discovery of protein based surfactants |
| Sectors | Agriculture Food and Drink Chemicals Manufacturing including Industrial Biotechology Pharmaceuticals and Medical Biotechnology |
| Description | Proteins are exceptionally versatile biomolecules. However, their use is impeded by their molecular complexity. This renders their selection and testing challenging and is generally trial-and-error. Our physics-embedded ranking method provides an excellent avenue for researchers and engineers to choose proteins with appropriate adsorption properties for use in lubrication, colloidal stabilization in food, pharmaceutical and biomaterials fields. With further developments and improvements, the tool can be used to replace synthetic polymers and help contribute to sustainable future. The research also has potential in developing a nascent field of physics embedded design of proteins as functional molecules, moving beyond statistical and evolutionary focus give thus far. |
| Sector | Agriculture, Food and Drink,Chemicals,Manufacturing, including Industrial Biotechology |
| Impact Types | Economic |
| Title | Protein Surface Adsorption Prediction |
| Description | This is related to the ML model that is developed to predict protein adsorption. The ML framework has a ranking tool that can be used to predict how a protein adsorbs to surface and compare it to a target protein that needs replacing. Therefore this model, when fully developed and expanded can be useful for researchers to use based on physical laws rather than black box prediction making protein selection and experimentation faster and more targeted. |
| Type Of Material | Computer model/algorithm |
| Year Produced | 2026 |
| Provided To Others? | No |
| Impact | No impacts yet. This is just developed and will be made public in the next quarter. |
| Description | Machine Learning Cross UK collaboration |
| Organisation | University of Edinburgh |
| Department | School of Informatics Edinburgh |
| Country | United Kingdom |
| Sector | Academic/University |
| PI Contribution | Our project team and Prof. Rik Sarkar at the School of Informatics collaborated on physics-embedded Machine Learning for protein adsorption to surfaces. . We used Lattice Self-consistent Field theory applied to 1000s of proteins for the first time to understand how proteins adsorb to hydrophobic surfaces on a large scale. This database provides a physics-embedded knowledge base. T |
| Collaborator Contribution | Prof. Rik Sarkar and our project team worked together with UoE on using the computational physics knowledge to establish a physics-embedded explainable ML to predict protein adsorption. Prof. Rik Sarkar helped fine tune our approach to ML, establishing the ML pipeline and helping execute the framework and improve our adsorption pipeline. Due to the explainable nature and simplicity of the ML models, the pipeline can enable discovery of novel proteins that can replace synthetic surfactants and lubricants and reduce trial-and-error experiments through informed selection. |
| Impact | https://doi.org/10.26434/chemrxiv.15000208/v1 |
| Start Year | 2025 |