Single-cell analysis

The single-cell analysis requires a volumetric nuclear stain image in Zarr format to find nuclei centroids. These nuclei centroids are used as seed points in a 3D watershed segmentation, providing morphological features. The nuclei centroids are also coordinates where to sample other channels for single-cell expression levels.

Convert to Zarr format

To facilitate parallel processing of volumetric datasets, the normalized TIFF stacks need to be broken into more manageable chunks. Fortunately, Zarr is a Python package that provides a chunk-compressed array data structure that we can use. To convert the TIFFs into a Zarr array, use the following command:

scout preprocess convert color0_rescaled/ syto.zarr -v -n 4

This will convert the TIFF stack in color0_rescaled into a Zarr NestedDirectoryStore called syto.zarr. The -n 4 argument specifies how many parallel workers to use in this conversion process. This command should be repeated for each channel.

Note: Using too many workers to convert TIFFs to Zarr may result in error messages about folder creation. A pull request has been submitted with Zarr regarding this issue.

Name each channel

First, it is convenient to provide the name of the staining target in a separate CSV file for building multiscale features later.

scout nuclei name sox2 tbr1 dn -o celltype_names.csv -v

Nuclei detection

The detection algorithm uses a filter based on curvature in nuclear signal, which is used later for segmentation. Also, the coordinates of each nucleus in physical space is also needed for future spatial analysis. These global positions in micron units can only be computed if the user provides the voxel size.

scout nuclei detect syto.zarr nuclei_probability.zarr centroids.npy --voxel-size voxel_size.csv --output-um centroids_um.npy -v

where nuclei_probability.zarr is a new Zarr array containing the probability of each voxel being a nucleus centroid and centroids.npy is a numpy array containing the voxel coordinates of the detected nuclei. In this case, we are also passing in –voxel-size voxel_size.csv in order to get a new numpy array –output-um centroids_um.npy containing the physical coordinates of the nuclei centroids.

Nuclei segmentation

To segment individual nuclei, run the following command:

scout nuclei segment nuclei_probability.zarr centroids.npy nuclei_foreground.zarr nuclei_binary.zarr -v

This will create two new Zarr arrays, nuclei_foreground.zarr and nuclei_binary.zarr. The nuclei_binary.zarr array contains a binary volumetric image where each nucleus as been separated with a watershed line. This watershed segmentation is used to compute morphological features for each nucleus with the following command:

scout nuclei morphology nuclei_binary.zarr centroids.npy nuclei_morphologies.csv -v

The result nuclei_morphologies.csv is a CSV file containing a table of morphological features for each detected cell.

Cytometric analysis

To sample fluorescence at each nucleus, run the following command:

scout nuclei fluorescence centroids.npy nuclei_fluorescence sox2.zarr/ tbr1.zarr/ -v

This command writes a few files in a new folder called nuclei_fluorescence, including a table of all mean fluorescence intensities (MFIs) for each cell. These MFIs can be used to gate cells using the following command:

scout nuclei gate nuclei_fluorescence/nuclei_mfis.npy nuclei_gating.npy 0.2 0.1 -p -v -r 1.5 1.5

where nuclei_gating.npy is a numpy array containing labels for each cell and gated channel. In this case, we are gating the first channel (sox2.zarr) at 0.2 and the second channel (tbr1.zarr) at 0.1. The flag -p plots a 2D histogram, and the -r 1.5 1.5 argument specifies the maximum range of the plot.