Complex gene regulatory networks define the fate of each cell in embryonic development. Understanding these mechanisms is a key challenge of developmental biology. Organisms with an invariant cell lineage such as C. elegans are a unique opportunity to study these networks on a single-cell level. Yet to gain statistically significant amounts of data requires lineages of many embryos. Despite substantial advances in the area of automated segmentation and tracking, existing approaches still require extensive human curation. To tackle this, we develop an approach to improve cell tracking performance over the state-of-the-art by employing deep learning to train a model using 4D convolutions. This allows to incorporate the temporal dimension into a deep learning approach for cell tracking. Our goal is to develop a tracking algorithm that is generic for different organisms and that can potentially be applied to organoids as well. The C. elegans specific knowledge of the cell lineage can be used to either (1) verify the results or (2) to use optimization for automated curation. Together with the Ercan lab at NYU that is focused on sex-specific gene regulation in C. elegans, we want to apply these tools to study two important biological processes that can only be addressed using a large number of cell lineages. Firstly, it is known that different sexes show significantly different morphologies. However, it is unclear at which point during development these begin to manifest. To address this, we will develop algorithms that automatically identify the sex and assemble two distinct probabilistic lineage trees that clearly highlight sex differences throughout development. Secondly, the loss of the dosage compensation machinery in hermaphrodites is lethal. However it is unknown why. Using our tracking tool, we will pin-point changes in the lineage that precede death. This will allow us to determine (1) if a specific alteration in development leads to death or if different alterations cause lethality in different animals and (2) at which point death can be predicted.
1. A. Krull*, P. Hirsch*, C. Rother, A. Schiffrin and C. Krull (2020) Artificial-intelligence-driven scanning probe microscopy. (*shared first) Commun Phys 3, 54. https://doi.org/10.1038/s42005-020-0317-3
1. P. Hirsch and D. Kainmueller. An Auxiliary Task for Learning Nuclei Segmentation in 3D Microscopy Images.arXiv:2002.02857. https://arxiv.org/abs/2002.02857
2. P. Hirsch*, L. Mais* and D. Kainmueller (*shared first). PatchPerPix for Instance Segmentation. arXiv:2001.07626. https://arxiv.org/abs/2001.07626
3. P. Hirsch and D. Kainmueller. An Auxiliary Loss for Learning Nuclei Segmentation in 3D Microscopy Images. Poster presentation at Frontiers in Imaging Science II, Janelia Research Campus, 1-4 May 2019.