Cellpose Demonstration
Introduction
Immunofluorescence experiments enable researchers to visualize expression of specific proteins in biological samples, like tissues or cell cultures, using antibodies conjugated to fluorophores to stain specific target proteins of interest. These samples are imaged using a fluorescence microscope to produce multi-channel images, with each channel typically representing an individual protein. Using image analysis techniques, the collective expression of the target proteins can be characterized to study disease mechanisms and infer biological pathways. A standard image analysis pipeline includes a step for image segmentation to identify biological structures within an image, like nuclei and cytoplasm. Cellpose is a popular software tool for segmentation of biological images and is included in the example Cellpose workflow on Mantle.
To import our example Cellpose workflow, click on the Import a template button on your Mantle dashboard and select Cellpose. This will import the required data types, example datasets, pipelines and runs, analysis environments, and analysis notebooks.
Raw data
Within the Mantle data lake, you can store images and associated metadata together as datasets. In our Cellpose demonstration workflow, a folder with the example images are stored as a generic image directory data type that contains the name of the dataset and the path to the directory. Data types can be created with additional metadata for the images, like biomarker for each channel, image resolution, and microscope used for acquisition.
In this template, images from the BBC014v1 dataset, available from the Broad Bioimage Benchmark Collection, are processed and analyzed.
Ljosa, V., Sokolnicki, K. L., & Carpenter, A. E. (2012). Annotated high-throughput microscopy image sets for validation. Nature Methods, 9(7), 637-637.
These images are taken from human MCF7 (human breast adenocarcinoma) and A549 (human alveolar basal epithelial) cell lines that are stained for 1) DNA and 2) NFκB. The cells are plated on a 96-well plate and each well was treated with one of 12 concentration points of TNFα. Higher concentrations of TNFα should induce cytoplasm to nucleus translocation of NFκB. 2-channel images were acquired from this experiment (channel 1: DNA, channel 2: NFκB) and uploaded to Mantle as a dir_images_generic dataset. The goal of this demonstration is to use Mantle to quantify and visualize the translocation of NFkB from these images.
Data processing
In our demonstration workflow, the images are processed using the cellpose-segmentation pipeline.
To run the pipeline, an input dataset needs to be specified along with parameters for running Cellpose. The input dataset can be any directory of multi-channel images. For our demonstration, the images are stored as the dir_images_generic data type. Our example dataset can be selected from the dropdown menu. Then, we need to enter parameters for running Cellpose. These parameters include:
Name of the pre-trained Cellpose model (e.g. “nuclei”, “cyto3”).
See the Cellpose documentation for more information.
Value for the channel with a nuclei marker.
0 = None, 1 = red, 2 = green, 3 = blue
Value for the channel with a cytoplasm marker.
For nuclei segmentation, this can be the same as “nuclei_channel” to load a single-channel grayscale image of just the nuclei channel.
0 = grayscale, 1 = red, 2 = green, 3 = blue
Expected pixel diameter of the objects being segmented.
“None” can be used for automatic estimation of the object diameter by Cellpose.
Running the Cellpose pipeline generates segmentation masks for each image from the input dataset. The masks are stored as a directory in the data type dir_masks_cellpose. Masks for the nuclei (BBBC014_nuclei) and the cytoplasm (BBBC014_cyto3) can be used with the original images to characterize biological features in the samples. In the next step, these files will be loaded into an analysis notebook to quantify the effect of TNFα on the spatial localization of NFκB in the cells.
Analyzing processed data
To analyze the images (BBBC014) and the segmentation masks (BBBC014_nuclei, BBBC014_cyto3), the Nucleus translocation of NFκB notebook is used. It is a Jupyter notebook that runs in the cellpose analysis environment.
Within the notebook, the Mantle SDK is used to download the directories of images and segmentation masks and mark the datasets as inputs to the analysis notebook. Then, the scikit-image package is used to characterize objects from the masks. Individual nuclei and cytoplasm are identified from the nuclei and cytoplasm masks, respectively. The NFκB channel from the images is used with the masks to quantify the expression of NFκB in each nucleus and cytoplasm within an image.
Nuclei and cytoplasm in each image are filtered to ensure that each nucleus is only associated with one cytoplasm and vice versa. Then the difference in intensity of NFkB for each cell is compared between the nuclei and the cytoplasm. The intensity difference is plotted against TNFα concentration on a dose response curve and there is a sigmoidal relationship between them. At high concentrations, there is a higher intensity of NFkB in the nuclei versus the cytoplasm. This indicates that there is translocation of NFkB to the nuclei in response to TNFα.
Wrapping up
Mantle is set up for storing, processing, and analyzing biological images. The Mantle data lake provides an easy way to store large imaging datasets with their corresponding metadata. Then image processing pipelines, like Cellpose, can be used to extract features from these images and analysis notebooks can turn these features into insights. Mantle is a streamlined platform that makes it easy to go from images to insights.