Two kinds of datasets are provided here. One is used for training and testing the deep learning model, another one is a ready-made cellular shape dataset for morphological analysis of C. elegans at single-cell level.

Experimental information

  • Transgenesis

    The membrane labeling strain carrying a single-copy transgene was generated using miniMos technique. Transgene insertion site was mapped using inverse PCR. The transgene insertion sites on chromosome are detailed in Supplementary Data 8.

  • Meta information

    There are 49 C. elegans embryos used in our work. The meta information, including time resolution, resolution, are detailed in Supplementary Data 1.

3D manual annotations for deep learning

  • Training dataset

    There are \(2\text{ (embryos) }\times 27\text{ (TPs in each embryo) }=54 \text{ }\)TPs semi-automatically segmented. They were firstly segmented with 3DMMS proposed by Cao JF, et al, and then checked by two experts to correct the segmentation errors. Each stack image with size \(205\times\times285\times134\) is saved as .nii.gz file, which can be loaded by NiBabel in python.

  • Evaluation dataset

    Another \(3\text{ (embryos) }\times 8\text{ (TPs in each embryo) }=24 \text{ }\)TPs are provided for the evaluation of the trained deep learning model. This dataset was obtained and saved as the way of that in the training dataset.

Normalized morphological atlas

  • Normalized segmentation

    In order to power the research on cellular morphology of C. elegans, we provided 17 embryos with cell-wise segmentation from 4- to 350-cell stages. The size and location variations are eliminated by considering the location of these 17 embryos plus another 29 embryos. Please note that CShaper failed to segmented some cells at later stages, which are listed in Supplementary Data 1.

Download the dataset

3D manual annotations (for deep learning) and morphological atlas can be download at the figshare repository. Please refer to the data description document for more details.