Pipeline Overview

Pipeline Architecture¶

The pipeline processes data through two parallel arms (the Cell Painting arm and the ISS arm) that operate independently on separate stacks of images before converging for final analysis in the standard workflow. The standard workflow is as follows:

Cell Painting Arm (Phenotype)¶

• Illumination Correction: Calculates a flat-field illumination correction to correct for uneven illumination.

• Illumination Application: Applies the flat-field illumination correction created in the previous step.

• Segmentation Check: Verifies cell/nuclei segmentation quality on a subset of images

• Stitch & Crop: Stitches fields of view into whole-well images and crops into tiles

Barcoding Arm (Genotype)¶

• Illumination Correction: Calculates a flat-field illumination correction to correct for uneven illumination.

• Illumination Application: Applies the flat-field illumination correction created in the previous step. Also performs inter-cycle alignment to align all subsequent barcoding cycles to cycle 1.

• Preprocessing: Compensates for spectral bleed-through, identifies barcode foci, and generates QC metrics

• Stitch & Crop: Stitches and crops to match Cell Painting tiles

Combined Analysis¶

Once both arms pass quality control, the final Analysis pipeline aligns Cell Painting and barcoding images, segments cells from the phenotypic stains, measures morphological features, reads an SBS barcodes for each SBS focus and selects a best match from the barcode library, and assigns barcode foci to cells — linking each cell’s genotype to its phenotype.

Entrypoints¶

The pipeline has three entrypoints, selected with Nextflow’s -entry option. These alter the sequence of image processing. The stich-align-crop workflow is as follows:

In STITCH_ALIGN_CROP, stitch, align and crop always run (they are not gated by qc_painting_passed / qc_barcoding_passed); only combined analysis is gated, and it requires both flags to be true. Relevant parameters: tileperside, final_tile_size, painting_scalingstring, barcoding_scalingstring, painting_channame, barcoding_channame, barcoding_shift_threshold, barcoding_corr_threshold, and combinedanalysis_include_barcoding_images (default false: only painting images, plus the optional barcoding FociObjects.tiff, are given to combined analysis). Output folders and QC report numbering differ for this entrypoint; see Outputs.

Set --check_duplicate_images true (any entrypoint) to run a pixel-duplicate check after illumination calculation/correction and preprocessing.

The “Stop-and-Check” Workflow¶

Making morphological measurements in high-content imaging data is computationally expensive. To avoid wasting resources on poor-quality data, the pipeline implements a “Stop-and-Check” workflow controlled by two parameters:

• --qc_painting_passed (default: false)

• --qc_barcoding_passed (default: false)

Phase 1 - Initial Processing: The pipeline runs through the QC checkpoints for both arms (SegCheck QC for painting, Preprocess QC for barcoding), generates QC montages, and stops before Stitch & Crop.

Phase 2 - Manual Review: You review QC outputs in results/workspace/qc_reports/.

Phase 3 - Production Run: If data looks good, set --qc_painting_passed true and --qc_barcoding_passed true, then resume with -resume. The pipeline continues from cached results.

Data Hierarchy¶

Understanding how the pipeline organizes data: