Figure 2. Optimisation of HCS assay using GW6604 and compound hits.

(A) Dose-response curves of TGF-β type I receptor kinase inhibitors LY-364947 (blue curves) and GW6604 (red curves) for four different parameters. The median IC₅₀ value of GW6604 was 1.9 μM and of LY-364947 was 0.2 μM. The parameters describe the texture of fibronectin signal within a colony (group of cells) (EntropyInten), the minimum distance to a neighboring nucleus (NeighborMinDist), the percentage of colonies with lower texture measurement of fibronectin signal within a colony than the threshold derived from a control population of colonies (ContrastCoocInten), and the ratio of convex hull to perimeter of a colony (ObjectConvexHullPerimeterRatio). (B) The scatter plot of two repeated runs of 7 plates (n = 2576) shows very good linear correlation with a Pearson coefficient of 0.927 for a single texture parameter (fibronectin signal). Negative control wells and inactive compounds scatter around the zero coordinate, whereas controls show phenotype strength-dependent scattering along the diagonal. (C) Table of false positive rate (FPR) and true positive rate (TPR) for the entire screen. Using the strict threshold of Euclidian distance >15 for the EMT cluster, the strong EMT inhibitor controls (GW6604: 3 and 10 μM) were identified as hits with a TPR of 98.3% and the negative control wells (DMSO) appeared as hits with a FPR of 0.21%. From the library 0.17% of wells were identified as hits. (D) Chemical structures of 16 compound hits. Simple heterocyclic and carboxylic acid analogs, steroidal and alkyl analogs, and extended multi-ring structures. Numbers correspond to the EPM IDs of each compound.