ESPCI web site
Decoding crowding dynamics in monolayers of motile bacteria using DiSTnet2D, a novel method for segmentation and tracking
Bacterial populations provide a compelling experimental model for studying active matter. Most studies have focused on systems driven by cell divisions (growth of microcolonies and biofilms) and by cell motility and hydrodynamical interactions (emergence of flocking and "active turbulence"). Here, I will present a system of "dry" active matter with a motile bacterium : dense monolayers of Pseudomonas aeruginosa.
I will first present DiSTNet2D, a novel method for object segmentation and tracking that leverages temporal context. DiSTNet2D is based on a Deep Neural Network that takes multiple consecutive frames as input to simultaneously perform segmentation and tracking. DiSTNet2D is hosted within Bacmman, an ImageJ plugin, which allows for generating training datasets, training, running the computation, and exporting the results.
Using DiSTNet2D, we analyze extensive fields of bacterial monolayers, achieving 100% accuracy in our observations across three decades of time. Our findings reveal complex dynamics : the wide cell size distribution, resulting from ongoing division, forms an amorphous structure that hampers crystallization. Additionally, we observe the spontaneous emergence of dynamic spatial heterogeneities and, at higher densities, characteristics indicative of a glass transition (specifically, a kinetic slowdown and the divergence of correlation times), which occur independently of individual cell motility patterns.