AI-Optimised Fermentation for Sustainable Protein Production from Food Side Streams

This international collaboration will develop AI methods for optimising fermentation processes that use food side streams as substrates to produce sustainable proteins for human consumption. It will be led by the University of Leeds (UoL) and the Commonwealth Scientific and Industrial Research Organisation (CSIRO) and includes four industrial partners from the agri-food sector.

Current alternative protein production often uses high-sugar substrates. This project aims to utilise food side streams as the fermentation substrate to increase sustainability and economic viability of protein production. Given the variability between different food side streams, AI will be used to aid optimisation of fermentation parameters like solids and moisture content, pre-treatment methods, pH levels, yeast strains, and nutrient supplementation.

Approximately one-third of food produced gets wasted. The project addresses this waste, contributes to creating a circular economy, while simultaneously producing protein to enhance food security for a growing world population. The project is aligned with two of the competition’s themes on AI in sustainable agriculture and food and AI to advance manufacturing and clean growth.

Each institution has complementary capabilities where the University of Leeds team offers experience in experimental data collection and preliminary model development and CSIRO brings expertise in scaling up these processes to ensure they have a broader, real-world impact. The partnership also addresses regional variations in food side streams enabling the development of adaptable models with greater generalisability. The projects 

activities are organised into five Work Packages (WPs):

• WP1 (UoL): Experimental data collection: Collection of data from fermentation processes using food side streams (short shelf-life soft fruit and Jerusalem artichokes) as substrates to produce protein.

• WP2 (UoL & CSIRO): Single side stream models: Leveraging the data from WP1, development of AI models to predict fermentation products (e.g., microbial dynamics, yield, and protein concentration) based on the side stream composition and fermentation parameters.
• WP3 (UoL): Adaptive modelling techniques: Using the data from WP1, development of transfer learning and Bayesian optimisation methodologies to significantly reduce the data collection burden for new side streams or equipment.

• WP4 (CSIRO): Scale up and out: The methodologies from WP3 will guide data collection from scaled-up fermentation trials at CSIRO using new food side streams aiming to maximise sustainability and economics.
• WP5 (UoL & CSIRO): Partnership and impact: Focusing on disseminating findings to the broader scientific community, sharing data, code, models, methodologies, and academic papers; and conducting partnership activities including visits, workshops, and training sessions.

The outcome of the partnership between UoL and CSIRO will be enhanced AI capabilities at each organisation, the career development of project team members and adaptable AI models that can be used by beneficiaries world-wide to efficiently optimise fermentation processes and assess the potential of new food side streams.