Skip to content

Quantification

Overview

The final step extracts features from the segmented objects and stores them in the AnnData table(s) within SpatialData objects.

  • Input: SpatialData objects containing segmentation masks and intensity images.
  • Output: AnnData tables containing single-cell feature matrices added to SpatialData objects.

Parameters of this step are defined in the Quantification part of configuration.

Supported features

  • Morphological Features: Geometric attributes such as area, centroid, eccentricity, and orientation. All skimage.measure.regionprops properties are supported.
  • Intensity Features: Statistical summaries (e.g., mean or median expression) for specified markers across the masked regions.
Feature Name Source / Implementation Description
mean Skimage: mean_intensity Standard intensity property from skimage.measure.regionprops.
max Skimage: max_intensity Standard intensity property from skimage.measure.regionprops.
min Skimage: min_intensity Standard intensity property from skimage.measure.regionprops.
median Custom: calculate_median Calculates the median intensity of the region.
sum Custom: calculate_sum Calculates the sum intensity of the region.
std Custom: calculate_std Calculates the standard deviation of intensity of the region.

Linking AnnData Tables to Segmentation Masks

In the SpatialData framework, every table should ideally be linked to a set of regions (masks). You can choose between two primary organizational strategies:

Separate quantification groups are defined to create distinct AnnData tables for each mask. * Example: A "cell" table linked to the cell mask and a "nucleus" table linked to the nucleus mask. * Benefit: Provides a strict 1:1 mapping between the table rows and the spatial labels.

Results from multiple masks are stored in a single AnnData table. * Example: Measurements for both "cell" and "nucleus" regions are stored in one table and linked to the "cell" segmentation mask. * Benefit: Simplifies downstream analysis by keeping all properties of a biological cell in one place.

Both approaches are supported; choose the one that best fits your downstream analysis workflow.

Execution

This module is fully parameterized via the configuration file and does not require manual intervention. Consequently, it can be executed via a Jupyter Notebook, standalone script, or as a component of the Nextflow pipeline; further details are available in the Execution Modes documentation.

Notebook workflow

The example Jupyter Notebook demonstrates the execution workflow using locally available data.

To complete the process, execute the following sections sequentially:

  • Read in config: Specify the path to the analysis configuration file and load the required settings.
  • Specify the overwriting strategy: Set the OVERWRITE_FLAG. If False, the pipeline will throw an error to prevent overwriting existing table. If True, existing table will be replaced. Use with caution!
  • Define the logger: Initialize the logging protocol to track execution progress and document the processing steps.
  • Define ROIs for processing: Identify the SpatialData objects to be processed. A demonstration cell is provided to truncate this list for faster testing.
  • Setup quantifiers: Initialize the quantifying objects defined in your configuration. These objects are created once and reused across all ROIs.
  • Run ROIs Quantification: Quantify the ROIs.

Script execution

python 05_quantify.py --exp_config ../examples/example_pipeline_config.yaml --overwrite