Challenge Provider: Hijos de Rivera Brewery (Maria Paredes, Jose Villarino)
Team: Long-Hung Pham, Shishun Liang, Ava Chan
Hackathon Mentors: Maria Paredes, Jose Vilariño. Pitch mentor: Dr Tatia Codreanu (I-X AI in Science)
The Challenge
The growing bioactive peptide market offers vast opportunities to create health-promoting ingredients from underused protein side streams. However, discovering and optimising these peptides remains challenging due to the complexity of protease–substrate interactions and the lack of predictive tools linking enzyme activity to peptide release.
The challenge aimed to develop an intelligent computational pipeline capable of predicting optimal, naturally occurring protease–substrate pairings to guide in silico proteolysis design. This would enable the efficient, scalable, and regulatory-compliant upcycling of low-cost protein sources into high-value ingredients with biofunctional properties such as antioxidant or anti-inflammatory activity.
The Approach
The team built an AI-driven discovery framework designed to model protease efficiency and predict cleavage site specificity across diverse protein substrates. Drawing on the MEROPS database of over 110,000 proteases and the UniProt database covering thousands of proteomes, the pipeline integrated enzyme–substrate information with bioactive peptide datasets from known crop species.
Using machine learning tools such as PanCleave alongside molecular docking software including HADDOCK and AutoDock, the model predicted protease–substrate interactions and prioritised likely bioactive peptide release sites. The workflow mined protein sequences to identify promising substrates, refined cleavage predictions through ML-based scoring, and applied molecular docking to assess enzyme–peptide binding stability. Together, these components formed a scalable proteolysis prediction pipeline that can accelerate discovery cycles and reduce experimental trial-and-error.
The Outcome
The resulting computational workflow demonstrated the feasibility of computer-aided proteolysis design for targeted peptide discovery. It provides a faster, data-driven alternative to conventional enzymatic screening, identifying novel proteolytic pathways with high commercial potential.
By combining protease activity modelling with bioinformatics and molecular simulation, the platform supports the upcycling of industrial protein by-products into functional, health-promoting peptides. The approach aligns with EU non-GMO regulations and offers a promising path to bring bioactive peptide innovation closer to scalable, sustainable production for food and nutraceutical applications.
Impact & Next Steps
The AI-driven peptide discovery pipeline accelerates the identification of bioactive peptides from industrial protein side streams, converting years of screening into rapid predictions that guide validation while enabling the upcycling of low-value by-products into high-value ingredients. Next steps involve identifying top peptide candidates by evaluating intrinsic activity and physiologically relevant abundance using molecular docking, molecular dynamics, and proteome-wide screening to pinpoint optimal sources.
These candidates will undergo experimental validation to confirm cleavage events and peptide release, followed by expansion of the bioactivity database, including antimicrobial, ACE-inhibitory, and immunomodulatory peptides and emerging substrates such as insects, algae, and cellular-agriculture proteins, and continuous model refinement informed by empirical data.