Patient-derived organoids have recapitulated in vivo physiological relevance, enabling a systematic evaluation of disease mechanisms and drug efficacies. Colorectal cancer organoids, specifically, featuring heterogeneous morphologies, have been inspected with image-based profiling. However, the association between morphological subtypes and functional parameters of the organoids is underexplored. Here, we reported two morphology subtypes which differentially responded to compound perturbations through image-based profiling. With object detection and image translation deep neural networks, the correlation of viability and apoptosis of the organoids and defined subtypes was elucidated. The underlying molecular heterogeneity of the organoids was also verified with scRNA sequencing. Additionally, we induced these subtypes with exogenous growth factors and a MEK inhibitor, and identified a relapse phenotype featuring intestinal stem cell signatures.
This code borrows heavily from NVIDIA/pix2pixHD and boutroslab/Supp_BetgeRindtorff_2021.