Tracking bacillus subtilis in 2D and inference of its run-and-tumble motility parameters

Abstract

System of motile bacteria can be considered as a living active matter because they act like energy consuming particles. Our rod-shaped bacterium, bacillus subtilis, uses its multiple flagella which are uniformly distributed over its body skin to swim though its environment. Its swimming motion can be roughly classified as two distinct phases, run and tumble. It is known that the bacterium escapes from the surroundings or relocates to the preferred environment by controlling the ratio of the two phases. In this work, we study this motility by comparing the statistics of swimming bacteria with that of run and tumble (RT) motion simulation. For the data collection, we adopt instance embedding, which is a deep learning (DL) based instance segmentation method, to detect the position of each bacterium from experimental video. The training dataset required for deep learning (DL) is prepared by generating synthetic images. Considering that the motility of bacteria in the real world entails various natural stochasticity and heterogeneity, we disperse the internal speed of active particles and include translational diffusion coefficient in the run-and-tumble equation of motion. As a result, our model successfully reproduces the probability distribution of relative displacement of the experimental trajectory, whereas the simulation with constant internal speed model results in much less variance in the horizontal directional distribution of the relative displacement compared to that of the experimental trajectory.