Study of the physiology of microorganisms: ESR4 (CHRH) has collected physiological data on different strains of Lactobacillus helveticus, one of the lactic acid bacteria (LAB) most used during the ripening of semi-hard cheeses such as Gouda and Cheddar. By studying L. Helveticus on synthetic media and developing a Genome-scale metabolic model (GEM), relevant pathways have been identified for the production of compounds related to flavour. ESR5 (INRAE) has studied the role of iron in the growth, interaction and production of volatile flavour compounds of bacteria from Livarot-type smear-ripened cheeses. By integrating multi-omics analyses and developing a metabolic model, this study provided a comprehensive framework for understanding microbial adaptation and cooperation on the cheese surface.
Study of fermented foods: ESR10 (INRAE) evaluated the impact of iron addition on the ripening characteristics of mini-Munster-type cheese by analysing multi-scale data collected throughout maturation. ESR12 (UM) did a comprehensive study of Maltese Ġbejna cheeses, focusing on microbiota identification by combining both molecular and morphological techniques and assessing their impact on cheese. The most promising LAB candidates were selected, and whole-genome sequencing was performed to assess their antifungal and anti-E. coli properties. ESR13 (NIZO) has set up a high-throughput screening system, and fermentations with 44 LAB were conducted in 6 different plant-based ingredients. Fermentation characteristics of each strain were studied for each strain-ingredient combination by the phenotypic data collection during fermentation. ESR14 (KUL) proposed a framework based on machine learning (ML) to interpret flavour data by using data-driven modelling techniques. ESR14 produced hybrid plant-dairy cheeses in a model system and analysed the contribution of LAB and plant proteins to the composition, proteolysis level, and flavour profile of these products.
Methods development for multi-omics integration: ESR6 (UT3) has proposed a new feature selection method to improve the interpretability of the kernel version of principal component analysis, and proposed a new R package, kpcaIG (10.32614/CRAN.package.kpcaIG). ESR6 and ESR8 (USZ) provided novel multiple kernel learning approaches for integrating diverse omics layers. ESR7 (UNIBO) has used classification models to infer the metabolic outcome from microbial gene expression profiles. Through a network diffusion algorithm, network modules were identified to elucidate antimicrobial resistance mechanisms, in collaboration with ESR12.
Integration of population-based concepts with constraint-based models: ESR1 (VUA) adapted and extended Community Flux Balance Analyses (
https://github.com/SystemsBioinformatics/dynamic-community-fba(öffnet in neuem Fenster) https://dynamic-community-fba.readthedocs.io/en/latest/(öffnet in neuem Fenster)) to simulate cooperative microbial consortia also in dynamic situations. ESR1 and ESR2 (CSIC) collaborated to integrate time-series metabolomics data into Flux Balance Analysis through dynamic modelling. ESR1 and ESR8 implemented the “Metabolic-Informed Neural Network”, a hybrid framework combining a Neural Network with GEMs to improve metabolic flux prediction from transcriptomics and proteomics data. ESR1 has set up a collaboration with ESR11 (KUL) to complement the insights gained about cooperative interactions with the previous methods, adding the effect of spatial structure through the use of Individual and Agent-Based Modelling. ESR2 has integrated population-based concepts with constraint-based models and developed a multi-scale modelling approach to describe community dynamics during cheese ripening. To deal with limited data availability, ESR8 tested SuperTML, a method that transforms tabular data into images and applies Convolutional Neural Networks with data augmentation to improve predictive performance in the context of data scarcity. ESR3 (CSIC) developed a computational generic framework for optimal and Model-Predictive Control of gene regulatory networks. Based on experimental results obtained by ESR10 regarding the effect of iron on various aspects of cheese, ESR3 developed a dynamic model to describe the growth of the microbial community involved in the cheese fermentation and ripening process.
Construction of Partial Differential Equation-based dynamic population models from individual-based models: ESR11 (KUL) has contributed to the field of Individual-based Model (IbM) in food microbiology through comprehensive methodological developments and practical applications by i) proposing a standardized conceptual framework to facilitate IbM implementations; ii) designing efficient numerical algorithms to accelerate simulations; iii) deriving simplified yet insightful Partial Differential Equation (PDE) representations to reduce model complexity; v) integrating multiscale datasets into the IbM framework, as well as leveraging IbM and PDEs to investigate microbial ecological phenomena.
Predictions of high-level properties with deep learning (DL) methods: ESR9 (USZ) project focused on predicting high-level properties in food fermentation using DL and ML methods. In collaboration with ESR13, ESR9 analysed datasets to uncover genetic factors associated with successful fermentations and developed ML and DL models to predict the fermentation performance of strains. Moreover, in collaboration with ESR7, ESR9 designed network-based approaches for feature selection.
Cheesemakers-oriented process modelling: ESR15 (INRAE) has proposed an innovative approach to studying phages in the dairy industry by developing a mechanistic model. This model has proven to accurately predict phage-bacteria interactions and paves the way for the development of decision support tools for cheesemakers, aiming to optimise their production processes.
