Machine learning-based estimation of spatial gene expression pattern during ESC-derived retinal organoid development

Abstract

Organoids, which can reproduce the complex tissue structures found in embryos, are revolutionizing basic research and regenerative medicine. In order to use organoids for research and medicine, it is necessary to assess the composition and arrangement of cell types within the organoid, i.e., spatial gene expression. However, current methods are invasive and require gene editing and immunostaining. In this study, we developed a non-invasive estimation method of spatial gene expression patterns using machine learning. A deep learning model with an encoder-decoder architecture was trained on paired datasets of phase-contrast and fluorescence images, and was applied to a retinal organoid derived from mouse embryonic stem cells, focusing on the master gene Rax (also called Rx), crucial for eye field development. This method successfully estimated spatially plausible fluorescent patterns with appropriate intensities, enabling the non-invasive, quantitative estimation of spatial gene expression patterns within each tissue. Thus, this method could lead to new avenues for evaluating spatial gene expression patterns across a wide range of biology and medicine fields.