Extended Data Fig. 2: Robust phenotypic signatures.
From: Phenotypic landscape of intestinal organoid regeneration
a, Top, phenotypic clusters and phenotypic groups (see Fig. 1b) identified using the multivariate feature array, and heat map of mean z-scored values per class for the indicated feature array. Aldob, aldolase B; Ar, area; D, DAPI; Ecc, eccentricity; II, integrated intensity; Lyz, lysozyme; MD, mass displacement; MI, mean intensity; SI, standard deviation of intensity; Sh, shape. n = 40,624, 34,265, 20,134, 31,687, 2,549, 702, 23,643, 48,992, 30,791, 21,342, 23,610, 32,773, 19,172, 41,931 and 29,939 organoids for classes 1–15 respectively. Bottom, phenotypic class assignment for organoids in control conditions; outer circle colour, phenotypic group; inner circle colour, assigned phenotypic cluster; n = 7367, 7982 and 33395 individual organoids for DAPT, CHIR99021 (CHIR) and DMSO conditions respectively. b, Potential loss of phenotypic resolution owing to phenotype grouping. Three phenotypic clusters are assigned to the ‘regenerative’ phenotype, but the exact 15-element phenotypic signature differs depending on whether PORCN or GSK3β is inhibited. Mean of n = 100, 4, 2 and 2 independent replicates for CHIR99021, Cpd2170, Cpd420 and Cpd1772 respectively. c, Top, relative z-score transformed abundances of 15 phenotypic clusters, shown as means ± s.d. Bottom, heat maps showing individual and mean phenotypic signatures for 50 replicates of the indicated control conditions. n = 50 sampled independent replicates for all 3 conditions. d, Left, randomization trial for reshuffling of phenotypic class labels in the entire dataset. Right, scatter plots and histograms showing the distribution of replica correlation and maximum class variance in permuted and non-permuted data for the selected and filtered conditions. Red dashed lines show the original selection stringency. e, Distribution of calculated P-values in the indicated groups of conditions. Top, scatter plot and kernel density estimation; bottom, box plots. Boxes, quartile range; whiskers, value interval with excluded outliers; solid lines, median values. n = 2,242 unique conditions corresponding to 5,204 individual wells. n = 279, 15 and 1,948 conditions for ‘hit’, ‘hit (one replica)’ and ‘not selected’ groups respectively. f, Top left, mean z-scored values per identified phenotypic class for the indicated feature array. Bottom left, heat map representation of the phenotypic signature for the indicated compounds, grouped by the most abundant phenotype in the original screen. Top right, tSNE map colour coded for the highest enriched phenotypic clusters. Every data point represents a compound. tSNEs were calculated from phenotypic signatures. Bottom, calculated reproducibility values per compound and numbers of replicates used for analysis of the indicated compounds. The replica count is the number of non-sparse (n > 10 organoids) independent wells recorded per condition. Assay performed in n = 8 independent replica from 2 independent organoid cultures. g, Flowchart depicting filtering of the screened conditions. h, tSNE maps colour coded for the highest enriched phenotypic clusters as indicated; every data point represents either a compound condition (top; n = 301 compound conditions) or a unique target gene-modality pair (bottom; n = 207 target genes); tSNEs calculated from 15-dimensional phenotypic signatures.
