Cardiac Fibrosis Rescue Screen Profiling

This repository contains the image analysis and image-based profiling pipeline for a cardiac fibroblast drug screen. It turns raw Cell Painting images into single-cell and bulk (well-level) morphology profiles for the 44 screen plates and the validation plate. This repository does not train models or call hits. The cardiac_fibrosis_rescue_screen_hit_calling repository uses the profiles for those analyses.

The screen

• 11 plate map layouts with 4 replicate plates each (44 plates in three batches), plus one validation plate
• 550 small molecule treatments and two controls: DMSO-treated failing and non-failing (healthy) cells
• A modified Cell Painting stain that swaps the RNA/nucleoli stain for F-actin, giving five channels: nuclei (d4), endoplasmic reticulum (d3), Golgi/plasma membrane (d2), mitochondria (d1), and F-actin (d0)

This plate map layout is the same for plates 1 through 10.

This plate map layout is specific to plate 11, which is a partial plate.

Pipeline

Each numbered module has a README with details and a bash script that runs it.

Module	What it does
0.download_data	Instructions for downloading the images
1.whole_image_qc	Flag over-saturated and blurry images with CellProfiler
2.illumination_correction	Correct uneven illumination and skip images that fail QC
3.cellprofiler_processing	Segment cells and extract morphology features with CellProfiler
4.image_based_profiling	Convert features to parquet, filter poor-quality cells, normalize, correct plate-position effects, and aggregate to single-cell and bulk profiles
5.validation_plate_profiling	Runs the same steps (illumination correction through bulk profiles) on the validation plate

Supporting folders:

• metadata: plate maps and barcodes, including treatment and pathway annotations
• utils: shared helper functions
• environments: conda environments

Environments

1. CellProfiler environment (fibrosis_cp_env): CellProfiler, for image QC, illumination correction, and feature extraction (modules 1, 2, 3, and the validation plate equivalents)
2. Preprocessing environment (fibrosis_preprocessing_env): pycytominer, CytoTable, and coSMicQC, for image-based profiling (module 4 and the validation plate profiling)

Create an environment with conda or mamba from the root of this repository:

mamba env create -f environments/preprocessing_env.yml

environments/hpc_create_envs.sh creates all environments on a Slurm cluster.

Outputs

The pipeline produces single-cell and bulk profiles, which module 4 documents in detail. Large intermediate files, such as images and parquet files, are not tracked by git.