Datasets for "EmbryoNet: Using deep learning to link embryonic phenotypes to signaling pathways"

This is the data repository of the training and test data sets for EmbryoNet. The data is structured in multiple packages. EmbryoNet_Models (DOI 10.48606/31) contains the trained neural networks, the other packages are imaging data. All data are brightfield timelapse images of one or multiple embryos recorded in multiwell plates in either the Acquifer Imaging Machine or the Keyence BZ-X810 microscope. The microscope type is included in the name of the archive, e.g. BMP_Acquifer.zip. Training data images are accompanied by json-files with the classification from human annotators, while test data sets also have the jsons of EmbryoNet's classification. The dataset EmbryoNet_Image-data: Stickleback 1 (DOI 10.48606/32) contains training data for the Stickleback version of EmbryoNet, and EmbryoNet_Test-data: Stickleback (DOI 10.48606/33) contains the evaluation data. EmbryoNet_Training-data: Medaka (DOI 10.48606/35) and EmbryoNet_Test-data: Medaka (DOI 10.48606/34) contain the respective data for Medaka. The other packages are zebrafish images. The two archives named EmbryoNet_Test-data 1&2 (DOI: 10.48606/29 & 10.48606/30) are the zebrafish test data sets. The zebrafish training data sets are named after the signaling molecule: EmbryoNet_training-data: BMP (DOI 10.48606/18), EmbryoNet_training-data: Retinoic acid (DOI 10.48606/20), EmbryoNet_training-data: Wnt (DOI 10.48606/21), EmbryoNet_training-data: FGF (DOI 10.48606/22), EmbryoNet_training-data: Nodal (DOI 10.48606/23), EmbryoNet_training-data: Shh (DOI 10.48606/25) and EmbryoNet_training-data: PCP (DOI 10.48606/26). EmbryoNet_training-data: WT (DOI 10.48606/16) contains the training data of untreated embryos. The datasets EmbryoNet_Training-data: Severities - Keyence (DOI 10.48606/28) and EmbryoNet_Training-data: Severities - Acquifer (DOI 10.48606/27) contain the training and evaluation data of the Severities experiments with different inhibitor concentrations. Inside a zip file the data is arranged in experiment folders, named in the format DATE_Molecule_concentration, e.g. 201222_FGF_10uM. Inside these experiment folders the data is organized after multiwell plate or microscope positions, A001-D006 for the Acquifer data and XY01-XY24 for the Keyence data.

本仓库为EmbryoNet模型训练与测试数据集的专属存储库。数据以多个数据包的形式进行组织。其中EmbryoNet_Models（DOI 10.48606/31）包含已训练完成的神经网络模型，其余数据包均为成像原始数据。所有数据均为明场延时成像图片，拍摄对象为多孔培养板（multiwell plates）中的单个或多个胚胎，成像设备涵盖Acquifer成像仪与基恩士（Keyence）BZ-X810显微镜。归档文件的命名中包含所使用的显微镜类型，例如BMP_Acquifer.zip。训练数据的图像文件附带由人工标注者生成的分类结果JSON（JavaScript Object Notation）文件，测试数据集则额外包含EmbryoNet模型自身输出的分类结果JSON文件。数据集EmbryoNet_Image-data: Stickleback 1（DOI 10.48606/32）包含EmbryoNet刺鱼版本的训练数据，而EmbryoNet_Test-data: Stickleback（DOI 10.48606/33）则包含对应的评估数据集。EmbryoNet_Training-data: Medaka（DOI 10.48606/35）与EmbryoNet_Test-data: Medaka（DOI 10.48606/34）则分别包含青鳉（Medaka）的对应训练与测试数据。其余数据包均为斑马鱼成像数据。名为EmbryoNet_Test-data 1&2（DOI: 10.48606/29 与 10.48606/30）的两个归档文件即为斑马鱼测试数据集。斑马鱼训练数据集以其所涉及的信号分子命名，具体包括：EmbryoNet_training-data: BMP（DOI 10.48606/18）、EmbryoNet_training-data: 视黄酸（Retinoic acid，DOI 10.48606/20）、EmbryoNet_training-data: Wnt（DOI 10.48606/21）、EmbryoNet_training-data: FGF（DOI 10.48606/22）、EmbryoNet_training-data: Nodal（DOI 10.48606/23）、EmbryoNet_training-data: Shh（DOI 10.48606/25）以及EmbryoNet_training-data: PCP（DOI 10.48606/26）。EmbryoNet_training-data: 野生型（WT，DOI 10.48606/16）包含未处理胚胎的训练数据。数据集EmbryoNet_Training-data: Severities - Keyence（DOI 10.48606/28）与EmbryoNet_Training-data: Severities - Acquifer（DOI 10.48606/27）则包含不同抑制剂浓度下的严重程度实验的训练与评估数据。在单个zip归档文件内部，数据按照实验文件夹进行组织，文件夹命名格式为DATE_Molecule_concentration，例如201222_FGF_10uM。每个实验文件夹内的数据则按照多孔培养板或显微镜视野位置进行整理：Acquifer设备的数据对应A001-D006，基恩士设备的数据则对应XY01-XY24。