Investigating Organelle Dynamics in Saccharomyces cerevisiae Using Advanced Imaging and Deep Learning

One of the overarching goals in quantitative eukaryotic cell biology is understanding how the cell orchestrates the dynamics of its organelles in pursuit of its physiological goals. Capturing how the suite of organelles interact with each other and with cellular-level dynamics has resisted both quantitative experiment and theory. Crucially, uncovering coordinated cellular and organelle dynamics requires quantitation of the multiple organelles residing within the same cell over time. Here we present our efforts at bringing high precision spatiotemporal fluorescence imaging to the scale of systems-level organelle dynamics in the model organism Saccharomyces cerevisiae. Our systems-level imaging strategy involves development of a deep learning pipeline to obtain spectrally dense information from spectrally coarse imaging. We trained a neural network to identify organelle signals from spectrally overlapping fluorescent labels imaged via widefield microscopy using hyperspectral confocal microscopy data as a source of ground truth of organelle numbers, sizes, and spatial positions within individual cells. We will highlight the utility of hyperspectral time-lapse imaging to uncover dynamic scaling and morphological relationships between organelles over the course of the cell cycle.