Full text
cellpaintr: Analyze CellProfiler Features in R Phatthamon Laphanuwat1, Caroline Ospelt1, and Christof Seiler123 1Center of Experimental Rheumatology, Department of Rheumatology, University Hospital Zurich, University of Zurich; 2Department of Advanced Computing Sciences, Maastricht University; 3Mathematics Centre Maastricht, Maastricht University Objectives •Leverage existing Bioconductor single-cell workflows for preprocessing, batch correction, data transformation, and visualization •Feature selection •Leverage prediction models Introduction •High-content imaging captures detailed morphological and functional features on a single-cell level •Widely used in drug screening assays and for generating large-scale image datasets for machine learning models •CellProfiler [1] implements a suite of image processing algorithms to segment cells and extract about 1,000 morphological and texture features •Analyzing CellProfiler features is tricky because features are highly correlated and hard to interpret Data •High-content imaging platform with a 7-color filter set (Thermo Scientific CellInsight CX7) •We adapted the Cell Painting staining protocol [2, 3] Figure 1: Fluorescent markers highlight the nucleus, endoplasmic reticulum, nucleolar, cytoplasmic RNA, actin cytoskeleton, Golgi apparatus, plasma membrane, and mitochondria. Methology segment morphological cells analysis my tinities cells cells cells hastaxislength sthefuture state cells cells control acells n scells Figure 2: CellProfiler implements image processsing algorithms to extract features. Learn state ngi sextract ce funtion serturbed feature f state a control celln predictor response matrix rectory PTI state ifatine pIPCcell.isperturbed fPcell is perturbed cell3 predictor predicted Pcell is perturbed matrix probabilities Figure 3: cellpaintr trains random forest classifiers on CellProfiler features. statecontrol cells can extract Effie pettitia feature cells cells can Icells effect log µ state perturbed Iii can cells cells cells get Mtrediated probability Figure 4: cellpaintr calculates perturbation effect sizes from predicted probabilities. Results Intensityall Intensity Texture RadialDistribution CorrCyto CorrER CorrMito CorrSYTO 0.0 2.5 5.0 7.5 10.0 0.0 0.5 1.0 1.5 log2 fold change −log10 p−value Target a a a a a IFN IL1 LPS PIC TNF Figure 5: Predict target vs. control with leave-one-out cross-valiation grouped by patient. Conclusions •Accessible for users familiar with single-cell RNA sequencing workflows •Grouping features in subgroups helps interpretation •Model free methodology → compatible with any prediction model Contact Information References [1] D. R. Stirling et al. Cellprofiler 4: improvements in speed, utility and usability. BMC Bioinformatics, 22(1):433, 2021. [2] M.-A. Bray et al. Cell painting, a high-content image-based assay for morphological profiling using multiplexed fluorescent dyes. Nature Protocols, 11(9):1757–1774, 2016. [3] P. Laphanuwat, E. Camarillo, C. Seiler, and C. Ospelt. Decoding synovial fibroblast morphology through cell painting imaging for drug screening applications. Annals of the Rheumatic Diseases, 84:511, 2025. EULAR 2025: European Congress of Rheumatology. Funding Digital Society Initiative Infrastructure & Lab of the University of Zurich
