Neural Recipes for Alt-Protein Food Products
Our innovation in the context of PLAN P aspires to remedy critical challenges faced by the food product design industry, and specifically the design of alt-protein products. The design and calibration of product parameters is a costly, time-consuming process, relying on the organisation, execution and validation of multiple experiments for trying different recipes. The challenge is magnified when the experimental design involves both ingredient and process characteristics, as is the case in PLAN P. The propose AI-driven solution alleviates these challenges faced by food product designers, by streamlining the training process and significantly lowering the volumes needed for training effective, accurate models for predicting the quality of an experimental design's results.
The proposed technological solution carries two major advantages in relation to currently existing solutions commonly applied to the problem of identifying the flavour and texture profiles of proposed product recipes. Firstly, it allows the prescriptive analysis of entire product creation chain, from protein and ingredient selection to treatment and fermentation techniques. Secondly, our AI technology is able to produce accurate predictions using drastically smaller training sets; In relation to traditional deep learning methods, it requires a training set reduced by a factor of 100, while presenting significantly better generalisation and transferability capacity. Therefore, it is less bound to a single product line, protein source, and final targeted product.
The alternative protein industry grows with leaps and bounds the last few years; still, it represents a miniscule fraction of the overall protein industry (meat, dairy, products using animal proteins). A transition to a more balanced market where alternative proteins occupy a sizeable share of the overall market is of critical importance, environmentally and financially, at the global scale. To do so, however, the industry has to move beyond its focus on health and environmental consciousness and address the perceived advantages of traditional protein products like appearance and texture to make alt-protein products equally desirable to a much broader audience. Nevertheless, alt-protein based food product modelling is a complex process, involving costly and time-consuming trial-and-error practices.
We use proprietary AI technologies to discover recipes for alternative-protein ingredients that provide high-quality tasting experience. The models operate over a broad spectrum of physiochemical and techno-functional properties of the materials used and processes applied for the manufacturing of the food product and predict the result of different combinations for key final products characteristics, including properties related to structural integrity and shelf life. The technology facilitates the quantification and approximation of the food product characteristics with their traditional, animal protein-based counterparts and thus eliminates the need of contacting a huge number of test trials for the exploratory analysis of different ingredients and processing steps, a huge cost centre for the delivery of tasty food products. Furthermore, we aspire to operationalise the solution as a no-code or small-code solution, allowing users to directly and timely benefit from our analytical capacities.
In the context of the PLAN P project, we have initiated the process of applying our AI solutions to the specific problem, taking advantage of the multi-disciplinary expertise and facilities brought in by project partners. The project constitutes an ideal testbed for deploying, testing and fine-tuning our approach for AI-driven recipe optimisation. It is expected that our innovation will reach a TRL of 6 at the end of the project. The foreseen next steps for our results is the conducting of a thorough market analysis, the formulation of a business plan and a go-to market strategy design and, finally, the deployment of a first full-fledged SaaS solution incorporating our technologies optimised for the domain.
