Abstract
C. elegans is one of the main model organisms used to study developmental biology. As it is transparent, and can be modified genetically to express fluorescent proteins, the growing embryo can be studied using 3D time-lapse images from a confocal microscope. This allows the cell lineage of C. elegans embryogenesis to be constructed at single-cell resolution, allowing the systematic observation, analysis and modeling of developmental processes. This thesis addresses, and improves on, methods to generate the C. elegans cell lineage automatically from 3D time-lapse images of embryo development and develops and uses methods to analyze obtained cell lineages based on spatial structure and gene expression.The first part of this thesis presents a novel cell nuclei segmentation method for 3D C. elegans embryonic time-lapse images, which significantly improves on the performance of the main system, StarryNite, currently used to track the C. elegans cell lineage automatically. We demonstrate the advantages of our bidirectional prediction based method over StarryNite. Several other modifications are also discussed.
The second part of this thesis presents a method, based on probabilistic relaxation labeling, to track C. elegans cells in 3D embryonic time-lapse images. The relative, instead of absolute, positions of the cells are used. Application to two embryo datasets shows that this method is highly accurate. Additionally, our algorithms can be implemented on multi-core CPUs, therefore the method improves both accuracy and speed for large-scale data analysis of 3D time-lapse images of live cells.
By using 90 wild-type 3D C. elegans embryonic time-lapse image datasets, we developed a C. elegans cell contact based reference model of the embryo in the third part of this thesis. Voronoi diagram based segmentation is used to predict the contact areas between cells during C. elegans embryogenesis. Five hundred ninety “effective” cell-cell contacts that have similar characteristics were extracted. This cell contact reference model was consistent with the known notch signaling related cell contacts and could be used to identify spatial variations in C. elegans embryos.
Co-clustering was explored in the fourth part of this thesis to detect correlated gene expression and cell fate in C. elegans cell lineage gene expression data. In the co-clustering approach of this thesis, a recently developed clustering algorithm, D-cluster, is applied to the left and right vectors in a singular vector space transformation of expression data. This co-clustering method was evaluated on a yeast gene expression dataset and on gene expression levels during C. elegans embryogenesis stages. The results indicate that cells with similar gene expression patterns in early embryogenesis tend to have similar roles in the adult.
Our work considerably improves the automatic generation of C. elegans cell lineages from 3D embryonic time-lapse images and provides effective computational tools for the analysis of C. elegans cell lineage data.
| Date of Award | 30 Aug 2016 |
|---|---|
| Original language | English |
| Awarding Institution |
|
| Supervisor | Hong YAN (Supervisor) |