Fig. 2. Organoid phenotype-profiles capture organoid viability.

a Representative example images of negative (0.1% DMSO) and positive control treated organoids (2.5 µM bortezomib, cyan = DNA, magenta = actin, yellow = cell permeability; representative images were selected and embedded in black background; scale bar: 50 µm). b Dose-dependent-trajectory of bortezomib drug effect. UMAP of organoid morphology at different bortezomib doses and (right panel) dose-dependent trajectory for three representative organoid lines. During the principle curve fitting, trajectory inference excluded cluster 4, a set of measurements representing mostly dead organoid particles comprising ca. 5% of all imaging data. c Dose-response relationship of organoid size and paclitaxel dose. D022T and D046T are highlighted as examples for responder/non-responder lines. Source data are provided as a Source Data file. d UMAP of organoid morphology highlighting D022T organoids treated at different concentrations of paclitaxel. e D046T organoids treated at different concentrations of paclitaxel. f Example images of D022T organoids treated with paclitaxel. cyan = DNA, magenta = actin, representative images of organoids were selected and embedded in black background; scale bar: 50 µm). g Association of organoid size of selected example organoid lines with viability determined by luminescence-based, ATP-dependent viability profiling with CellTiter-Glo (CTG), which was performed in parallel with imaging on a subset of drug treatments for benchmarking. h UMAP visualization of viability predictions for organoids within our dataset, based on supervised machine learning of organoid viability using classifiers trained on positive (high-dose bortezomib and SN-38) and negative (DMSO) controls (live-dead classifiers, LDC). i Association of LDC and example organoid features (size, DAPI, actin and permeability dye intensities) with benchmark CTG viability read out. Source data are provided as a Source Data file.