A SEMI-AUTOMATIC MODELLING APPROACH FOR THE PRODUCTION AND FREEZE-DRYING OF LACTIC ACID BACTERIA Thomas Chabin, Marc Barnab´ e, Alberto Tonda, Nadia Boukhelifa, Fernanda Fonseca Eric Dugat-Bony, H´ el` ene Velly, Evelyne Lutton, Nathalie M´ ejean Perrot UMR 782 GMPA, INRA 1 av. Lucien Br´ etigni` eres, 78850, Thiverval-Grignon, France email: {firstname.lastname}@inra.fr KEYWORDS Complex Systems, Multiscale Modelling, Interactive Modelling, Optimisation, Expert Knowledge, Visualisa- tion, Freeze-Drying of Lactic Acid Bacteria. ABSTRACT The production system of freeze-dried lactic acid bacte- ria involves several processes, but its impact on bacteria resistance is still not well understood. This system can be defined as a complex one since it depends on mul- tiple scales: the Genomic, the Cellular and the Popu- lation scale. The scarcity of data available for building models leads us to propose an approach that makes use of expert knowledge. In this paper we present a semi- automatic modelling tool, LIDEOGRAM and discuss how it contributes to insight formulation and rapid hy- pothesis testing. New results show that LIDEOGRAM is able to produce more robust modelling hypotheses when experts can interact and revisit the genomic data preprocessing. INTRODUCTION Complex systems are found in many features of nature and science, ranging from the economic and social struc- ture of a city to the global climate, or from the be- haviour of a single cell to the behaviour of the intricate interactions of the human brain. Complex systems in- volve numerous components linked by non-trivial rela- tionships, and are challenging to study. Modelling such complex systems, by summarising available knowledge into a mathematical or computational representation, is not a trivial task. Concentrates of Lactic Acid Bacteria (LAB) are widely used in the food industry for manufacturing products such as yoghurt, cheese, fermented meat, vegetables and fruit beverages. The production of freeze-dried LAB is a complex food system due to its multi-scale and multi- step properties. One of the main challenges that need to be tackled ahead is to understand the origin of the LABs resistance and/or their sensitivity to the whole production system. Models describing parts of the sys- tem for a specific strain of bacteria are found in the literature (Passot et al. (2011)). However, to the best of our knowledge, no models have been proposed to repre- sent the whole process. Automatic modelling approaches have already been pro- posed for complex systems such as metabolic networks (Schmidt et al. (2011)), or for various multi-scale pro- cesses (Hasenauer et al. (2015)). These approaches often require a significant amount of data, however, gathering data on the freeze-drying process of LAB is expensive and time-consuming. Little amount of data is thus gen- erally available, and this is a major issue for automatic modelling. To compensate for the lack of data, expert knowledge on the process can be exploited. We show in this paper how such knowledge can be integrated within a modelling process, based on a semi-automatic scheme. The paper is organised as follows: First we present some background on complex systems, on semi-automatic modelling and on expert knowledge integration. The target system and the dataset are then described. Next we detail our semi-automatic modelling software and show some experimental results. Finally, results are dis- cussed and conclusions are drawn. BACKGROUND Complex Systems in Biology and Food Systems: Expert Knowledge Integration Methods A complex system is defined as a system made of mul- tiple processes, entities, and nested subsystems. Global properties emerge through a series of phenomena occur- ring at different scales (Ladyman et al. (2013)). Ap- propriate descriptions with high expressiveness and lit- tle uncertainty of the underlying mechanisms is needed to elucidate such systems. Building models of complex systems is crucial, but highly difficult. It usually re- quires a robust framework, with strong iterative inter- action combining computational intensive methods, for- mal reasoning and experts from different fields. In such context, optimisation plays an important role (Lutton et al. (2016)). Properties of food systems (such as uncer- tainty and variability, heterogeneity of data, coexistence of qualitative and quantitative information, conjunction of different perspectives) raise the focus on another es-