Segmented Gel Electrophoresis Images used for Training GelGenie Models

This repository contains gel images and corresponding hand-labelled segmentation maps (575 total) from various sources. The images have various sizes, shapes, gel contents and imaging conditions. In more detail: The matthew_gels (301 total), matthew_gels_2 (85 total) and nathan_gels (37 total) folders contain images originating from previous experiments in the Dunn Lab. The quantitation_ladder_gels (35 total) were generated specifically for the GelGenie project and contain gel ladders in each well. These were used to generate the data for Figure 1 in the main paper. The stella_gels_for_finetuning (26 total) folder contains gel images provided to us from Siyuan Stella Wang (Wyss Institute/Dana-Farber Cancer Institute) which we used to finetune our U-Net model and produce the results in Figure 4C of the main paper. The external_gels (25 total) images were gifted to us from 5 different researchers: Yichen Zhao (images generated at the University of Waterloo), Huangchen Cui (images generated at Tsinghua University), Thomas Mayer (images generated at the Technical University of Munich), Joana Reis (images generated at the Dana-Farber Cancer Institute) and Ricarda Törner (images generated at the Dana-Farber Cancer Institute). These were used as an external unseen test set for our fine-tuned model. The lsdb_gels (66 total) folder contains segmentation maps of images downloaded from the RGB Caps dataset, which is available from https://dbarchive.biosciencedbc.jp/en/rgp-caps/desc.html. The original images have been shared with a Creative Commons Attribution-Share Alike 2.1 Japan license. Permission was obtained from the data depositors for the training of our models on these images, as well as for the sharing of derived segmentation masks using the CC-BY license. To use this portion of the dataset, you will need to download the original images from the website, and place them alongside our segmentation masks, mirroring the same setup used for all the other datasets in this repository. The folders are organised as follows: The images, val_images and test_images contain the original images, split into training/validation/testing partitions respectively (except for the lsdb_gels dataset). The masks, val_masks and test_masks contain the corresponding segmentation maps for each image in the main folders (they have identical naming i.e. an image named test_image.tif will have a segmentation map labelled test_image.tif). The external_gels folder does not have validation or test partitions. The segmentation maps are 8-bit images, for which a white pixel (0) corresponds to a background pixel and a brown pixel (1) corresponds to a foreground pixel. This is the main dataset used to train the GelGenie models, for which the corresponding source code and GUI can be downloaded at https://github.com/mattaq31/GelGenie.本仓库包含来自不同来源的凝胶图像（gel images）与对应的人工标注分割掩码（hand-labelled segmentation maps），总计575份。这些图像在尺寸、形态、凝胶组分及成像条件上均存在差异。详细说明如下： matthew_gels（总计301份）、matthew_gels_2（总计85份）与nathan_gels（总计37份）文件夹内的图像均来自邓恩实验室（Dunn Lab）既往实验。 quantitation_ladder_gels（总计35份）专为GelGenie项目构建，每个泳道均包含凝胶分子量标准（gel ladder），该数据集用于生成主论文中图1的相关数据。 stella_gels_for_finetuning（总计26份）文件夹内的凝胶图像由Siyuan Stella Wang（怀斯研究所/Wyss Institute/丹娜法伯癌症研究院）提供，我们使用该数据集对U-Net模型进行微调，并生成了主论文图4C中的实验结果。 external_gels（总计25份）由5位研究者赠予：Yichen Zhao（滑铁卢大学生成）、崔黄晨（清华大学生成）、Thomas Mayer（慕尼黑工业大学生成）、Joana Reis（丹娜法伯癌症研究院生成）与Ricarda Törner（丹娜法伯癌症研究院生成）。该数据集被用作微调后模型的外部未知测试集。 lsdb_gels（总计66份）文件夹内包含从RGB Caps数据集下载的图像的分割掩码，RGB Caps数据集可从https://dbarchive.biosciencedbc.jp/en/rgp-caps/desc.html获取。原始图像采用知识共享署名-相同方式共享2.1日本版（Creative Commons Attribution-Share Alike 2.1 Japan）许可协议发布。我们已获得数据存储者的许可，可使用这些图像训练模型，并以CC-BY许可协议分享衍生的分割掩码。若要使用本数据集的该部分，您需要从上述网站下载原始图像，并将其与我们提供的分割掩码按本仓库内其他数据集一致的目录结构放置。 本仓库的文件夹组织方式如下： images、val_images与test_images文件夹分别存放原始图像，并划分为训练集/验证集/测试集分区（lsdb_gels数据集除外）。 masks、val_masks与test_masks文件夹分别存放上述主文件夹内图像对应的分割掩码，二者命名完全一致（例如名为test_image.tif的图像，其分割掩码文件亦为test_image.tif）。 external_gels文件夹未划分验证集与测试集。 本数据集的分割掩码均为8位图像，其中像素值为0的白色像素代表背景，像素值为1的棕色像素代表前景。 本数据集是训练GelGenie模型的核心数据集，对应的源代码与图形用户界面（GUI）可从https://github.com/mattaq31/GelGenie下载。