Microbial communities (microbiomes) drive critical processes from plant nutrition and crop productivity to biogeochemical cycling and human health. Understanding and engineering microbiomes represents a major challenge for sustainable agriculture, medicine and climate resilience. Yet microbiomes remain poorly predictable because they are highly diverse, strongly dynamic across space and time, and supported by an immense metabolic repertoire. These properties generate a combinatorial explosion of microbial and metabolic interactions. Environmental or host drivers often explain only a modest fraction of compositional variance, leaving microbial interactions as a central, under-characterized determinant of assembly. However, microbiomes are experimentally tractable: fast generation times, culturable on diverse media, fluorescent tagging and high-throughput approaches for measuring growth (millifluidics, automated imaging, multi-omics) that can enable large and time-resolved datasets amenable to mechanistic inference. However, current models are fragmented: coarse-grained community models capture large patterns but lack metabolic functioning and spatial resolution, while genomescale models reveal mechanisms in small systems but do not scale. Mechanistic and predictive descriptions of how microbial interactions depend on diversity, space and function are largely missing. The goal of the MIMMIC project is to develop new mathematical models tightly coupled with reductionist experiments to provide a mechanistic understanding of the microbial interactions underlying microbiome assembly dynamics across diversity, spatial and functional scales. The project is structured around 4 work packages (WPs): in WP1, we will combine full-factorial synthetic bacterial community experiments in millifluidic systems with stochastic dynamical models, higher order interaction terms and information-theoretic criteria to quantify how community composition and diversity reshape interaction networks, and to identify the minimal model structure needed to predict assembly across diversity gradients. In WP2, we will use the Petri dish as a model ecosystem and infer spatial interactions in synthetic communities from multimodal time-lapse imaging (morphology, speckle activity) via spatio-temporal graphs, optimal design theory and Physics-Informed Neural Networks. In WP3, we will connect genome-scale metabolism to community dynamics by generating time series of synthetic communities combining RB-TnSeq functional genomics, time-resolved profiling of extracellular metabolites (exometabolomics), and community-wide gene expression measurements (metatranscriptomics). These data will be integrated into genome- and phenome-structured dynamical models to infer how genes, metabolic dynamics and interaction phenotypes co-vary across synthetic communities. WP4 will ensure project management, communication and dissemination. This project will be performed by a multidisciplinary consortium gathering 5 research units covering diverse expertise: mathematical modeling applied to microbiomes (BIOGECO, MAIAGE, ISA), microbial ecology and functional genomics (IRHS), and systems biology (TBI).
The project will advance theoretical ecology by clarifying higher-order, context-dependent interactions, enable more reliable microbiome engineering and produce broadly applicable methods for inverse problems: image-based inference and multiscale modeling. More specifically, it will contribute novel mathematical tools for (i) the integration of complex biological interactions in dynamical systems, (ii) inference of higher-order, metabolic and spatial interaction laws from incomplete data, and (iii) hybrid mechanistic/data-driven modeling that links genes, metabolites, interaction phenotypes and community dynamics.