Unravelling the Mechanisms Invisible to the Naked Eye – Inferring Complex Microbial Communication through Graph-Based Deep Learning

Single-cell methods have enabled deciphering the role of individual cells within for example ontogenesis or diseases like cancer [2, 3, 4]. In particular, with the aid of public databases, it is possible to infer cell-cell communication (CCC) from expression of ligands and receptors at single-cell resolution [1]. However, these methods primarily focus on humans and mammalians. As such, methods for obtaining and analyzing bacterial single-cell data so far do not go beyond proof-of-concept studies (though one partial exception is [5]). This is especially true for microbial CCC. In such diverse, unicellular populations, communication events are driven by more varied mechanisms than in humans, for example via metabolite exchange (cross-feeding) or quorum sensing. As a consequence, dedicated methods for bacterial single-cell data are required.

Here, we propose to develop a CCC inference method tailored towards bacteria, aiming to uncover potential interactions at single-cell resolution. Our approach will integrate genomic, transcriptomic, and metabolomic data from microbial samples coupled with public knowledge drawn from databases. First, we will focus on metabolite exchange and then expand our methodology later with more complex capabilities. Lastly, we will introduce our use case, where we will apply our approach on fermented foods, aiming to discover interactions driving flavour and texture development.