Machine Learning and network based methods for the analysis of biological systems involved in food processing

Throughout this thesis, we used network-based approaches such as network diffusion, community detection and centrality measures to successfully identify new genes related to E. coli’s antimicrobial resistance to widely used antimicrobial’s. Interestingly, some of these inferred genes were also experimentally validated. On the other hand, network-based centrality measures, combined with the dynamic constraint-based modeling approaches applied to genome scale metabolic models has permitted the development of a new tool used to construct context-specific metabolic networks in yeast cells. Alongside, we also considered applying machine learning methods to analyze cheese-related data collected by our colleagues within the E-MUSE consortium. We demonstrated a strong microbiome-metabolome relationship in cheese, by using a microbial metatranscriptome dataset for training the predictive model and another one as independent validation, and made use of it to estimate the cheese flavor profiles directly from the microbial gene expression. The accuracy of the trained models could reach 50 to 83 %. Moreover, the analysis of the genes selected by the modeling procedure showed their consistency with biological pathways, mainly the metabolic ones associated to flavor production.