buscar
Hit calling — identifies biologically active perturbations from single-cell morphological profiles using distribution-level scoring.
Experimental Tools
Next-generation tools under active development in the WayScience organization, designed to become the foundation of Cytomining v2.
What does Cytomining v2 solve?
The current Cytomining stack was designed around 2D single-cell data from CellProfiler. As the field moves toward 3D organoid imaging, larger-scale archives, and deep learning feature extraction, several gaps have emerged: no standardized image catalog, images and features stored separately, no 3D support, and hit calling that collapses single-cell heterogeneity. The tools below are purpose-built to close each of these gaps — together forming a fully traceable, format-agnostic, 3D-capable profiling pipeline.
Tools
iceberg-bioimage
Data cataloging — scans bioimaging stores and publishes image metadata to Cytomining-compatible Parquet warehouses via Apache Iceberg.
OME-arrow
Image storage — stores microscopy images alongside metadata and derived data in a unified, queryable Apache Arrow format.
zedprofiler
3D feature extraction — extracts morphological features from volumetric microscopy images for CPU-efficient high-content profiling.
Cytomining v2 pipelines
Standard (2D)
🔬 Raw Images
iceberg-bioimage
catalogOME-arrow
store📊 Feature Extraction
extract
CytoTable
harmonize
coSMicQC
QC
pycytominer
processbuscar
hit call3D Organoid
🔬 Raw Images
iceberg-bioimage
catalogOME-arrow
storezedprofiler 3D
3D extract
→CytoTable
harmonizecoSMicQC
QCpycytominer
processbuscar
hit callYellow = new 3D-capable step. Purple = new data infrastructure. Blue = existing Cytomining tools.
What each tool solves
🗄️
Problem: Raw bioimaging archives have no standard catalog — finding, versioning, and joining images to downstream data requires bespoke scripts per lab. Solution: Scans any image store into a versioned Apache Iceberg catalog that directly exports Cytomining-compatible Parquet warehouses.
🏹
Problem: Images and feature tables live in separate systems — linking a numeric outlier back to its source cell requires error-prone manual joins across formats. Solution: Embeds images as first-class columns in Arrow tables so features, metadata, and pixel data travel together and can be queried or exported as tensors.
📦
Problem: Classical profiling tools only extract 2D features — organoid, cleared-tissue, and confocal z-stack experiments are left without a first-class CPU-efficient feature extractor. Solution: Extracts morphological features directly from 3D volumetric images with anisotropic spacing correction, no GPU required.
🔍
Problem: Population-level hit calling averages away biologically meaningful cell-to-cell variation — heterogeneous responses and rare subpopulations are invisible to copairs-style metrics. Solution: Scores perturbation efficacy and specificity directly on single-cell distributions using Earth Mover's Distance, preserving heterogeneity throughout hit calling.