A codebase for automated analysis of High Content Screens.

The input image-data are assumed to be generated from High Content Screen plates, with multiple fields of view acquired per well, and 3 channels recorded per field, corresponding to nuclei, cells, and aggregates. For a 384-well plate with 9 fields per well, 10'368 images are acquired in total per plate which are processed by the aSynAggreCount package to quantify aggregate-positive cells. For each image, aSynAggreCount performs image pre-processing to correct for the presence of uneven illumination and digitization noise, then performs segmentation of the structures of interest (nuclei, cells, aggregates), and finally applies co-localization analysis to characterize the presence of aggregates inside cells. An illustration of the image processing pipeline is shown below.

The segmentation of nuclei and cells is performed with pre-trained Deep Learning models (see section Modules below). Aggregates are segmented using a sequence of conventional image processing filters, calibrated for the datasets of our study. An illustration of the segmentation quality is shown in the figure below where white lines correspond to the segmentation and blue to the original image intensities for nuclei (A), cells (B) and aggregates (C).

aSynAggreCount utilizes the segmented data to perform co-localization analysis and characterize aggregate inclusions in cells by computing various quantities of interest (QoI), such as the percentage of aggregate-positive cells and the number of detected aggregates per cell. The figure below shows a visualization of these QoI.

Create and activate a new conda environment

conda create --name aSynAggreCount python=3.9.18
conda activate aSynAggreCount

• Installation instructions
• Summary (for Linux):

# For GPU users:
pip install --extra-index-url https://pypi.nvidia.com tensorrt-bindings==8.6.1 tensorrt-libs==8.6.1
pip install -U tensorflow[and-cuda]
# Verification of the installation:
python3 -c "import tensorflow as tf; print(tf.config.list_physical_devices('GPU'))"

• Installation instructions
• Summary:

pip install stardist

• Installation instructions
• Summary:

pip install cellpose[gui]

The following packages are required to run the code: scikit-image matplotlib click pytest pyyaml pandas plotly kaleido.
They can be installed as follows:

pip install -r requirements_XXX.txt

where XXX is the file suffix corresponding to your OS: requirements_Ubuntu22.04.txt or requirements_MaxOSX.txt.

Nuclei are segmented using the StarDist pre-trained deep neural network.
Note: Results improve if Background Equalization is done before segmentation.

Two methods are implemented to segment cells:

• The Cellpose pre-trained deep neural network.
• The distance-intensity algorithm, inspired by CellProfiler's Propagation method.
  • Identifies cell boundary as a Secondary object.
  • Uses primary objects seeds.
  • Secondary objects are identified based on the shortest path to an adjacent primary object.
  • The distance metric is a sum of the absolute differences in a 3x3 (8-connected) kernel.
  • The dividing lines between secondary objects are determined by a combination of: the distance to the nearest primary object, and intensity gradients.

Aggregates are segmented through a sequence of conventional Image Processing filters, with their parameters calibrated to our data.

Parameter adjustment:
Before running the analysis on your data you need to first adjust the image segmentation parameters, inside the user configuration file: applications/setup.yml.

Run:
After adjusting the contents of applications/setup.yml to your data, run the analysis as follows:

python main.py

The code is developed by Athena Economides. The corresponding publication is in preparation.

If you are interested to use this code for your analysis, please contact via email.

Athena Economides, PhD
Lab of Prof. Adriano Aguzzi
Institute of Neuropathology
University of Zurich
Schmelzbergstrasse 12
CH-8091 Zurich
Switzerland