Extended Data Fig. 1: Reproducibility of the image-based screen.

a, An example image (MIP of a confocal z-stack) of an organoid, with annotated regions along the major axis, and showing antibody staining for lysozyme (Lyz) and aldolase B (Aldob). Lysozyme-positive Paneth cells are indicated by white arrowheads. Scale bar, 50 μm. b, Left, an example image of a stitched well-level overview. Scale bar, 100 μm. Middle, enlarged regions from the left panel. Scale bar, 100 μm. Right, segmentation masks, with object unique identifier numbers overlaid. a, b, Representative images of the DMSO control condition, repeated in n = 471 independent replicates with similar results. c, Top, reproducibility of detected organoid counts in replicas of the library plates. Heat map of Pearson’s correlation of organoid counts in screen replicates, with plate replicates grouped by parent library plates as indicated; n = 384 wells per plate. Bottom, heat map of standard deviation of organoid size (z-scored organoid area) for respective well positions. Passive and active controls are in wells outlined by the white box. n = 18 independent plates. d, Left, covariance matrix of the 34 extracted features, clustered by covariance (c-means clustering, ten clusters). Ch, channel; StdIntensity, standard deviation of intensity. Right, heat map showing principal component loading of the extracted features; bar chart above, eigenvalues of first ten principal components. e, Top, changes in the accuracy of the naive Bayes predictor throughout the backwards-feature-elimination test. Bottom, final set of nine features (‘Algorithm-selected’) and list of features used as input for clustering with PhenoGraph. f, Distribution of mean values per well of the indicated features in active and passive control conditions. CHIR is the GSK3B inhibitor CHIR99021; DAPT is a γ-secretase inhibitor. Centre line, median; boxes, quartile range; individual data points overlaid. Two-sided t-test, with P-values shown in the plot. MeanInt, mean intensity. n = 471, 100 and 101 independent samples for the DMSO, CHIR99021 and DAPT conditions respectively. g, Z-scored organoid features used for clustering, shown for individual well positions of in-plate replicates of active and passive control conditions. n = 6, 6 and 20 wells per plate for DAPT, CHIR and DMSO respectively; shown are mean values over n = 18 assay plates. D, DAPI; Eccentr., eccentricity, Integr. Int., integrated intensity; Mass disp., mass displacement for the indicated immunofluorescence channels; Std Int., standard deviation of intensity. h, Mean z-scored values for the indicated features per well from the entire dataset, sorted in ascending order. Every data point is an average value per well, with conditions colour-coded as in the key. n = 5,250 independent samples measured in the screen. i, tSNE maps colour coded for z-scored values of indicated features (left to right: eccentricity, organoid area, mean intensity of aldolase B and mean intensity of lysozyme antibody staining) and PhenoGraph classification. Every data point represents an individual organoid, with sampling stratified by PhenoGraph assignment. The sample size is 5,000 individual organoids.

Source data