SUPERVISED MACHINE LEARNING FOR THE QUANTIFICATION OF LIPID DROPLETS DYNAMICS

Lipid droplets (LDs) are essential cellular organelles due to ability to accumulate and store lipids. LDs dynamics are associated with various cellular and metabolic processes. Accurate monitoring of LD`s size and shape is of prime importance as it indicates the metabolic status of the cells. Unintrusive continuous quantification techniques have a clear advantage in analyzing LDs as they measure and monitor the cells` metabolic function and droplets over time. Here, we present a novel machine-learning-based method for LDs analysis by segmentation phase-contrast images of differentiated adipocytes (in vitro) and adipose tissue (in vivo). We developed a new workflow based on the ImageJ WEKA segmentation plugin, which provides an accurate, label-free, and live single-cell and organelle quantification of LD-related parameters. By applying the new method on differentiating 3T3-L1 cells, the size of LDs was analyzed over time in differentiated adipocytes and their correlation with other morphological parameters.

Moreover, we analyzed the LDs dynamics during metabolic changes such as lipolysis and demonstrated its ability to identify different cellular subpopulations based on their structural, numerical, and spatial variability. This analysis was also implemented on unstained ex-vivo adipose tissues to measure adipocyte size, an important readout of the tissue`s metabolism. The presented approach can be implemented in different LD-related metabolic conditions and lead to a better understanding of lipid droplets` biogenesis and function in vivo and in vitro while serving as a new platform allowing for short and accurate datasets screening.