Virtual tissue microstructure reconstruction by deep learning and fluorescence microscopy

Analysis of tissue microstructure is essential for understanding complex biological systems in different species. Studying the three-dimensional (3D) microstructure of tissues, such as the liver, is particularly fascinating due to its essential roles in metabolic processes and detoxification. The liver, composed of diverse cell types like hepatocytes, stellate cells, endothelial cells, and immune cells, exhibits a highly organized structure. Emerging technologies such as 3D microscopy and fluorescence microscopy offer exciting new possibilities for such analysis. Creating digital models that faithfully represent the inherent 3D architecture requires simultaneous imaging of multiple markers across large volumes, hindered by limitations such as poor antibody penetration, restrictions on fluorescent markers, and time-intensive procedures. Here, we present a generic approach for virtual 3D tissue microstructure reconstruction through the integration of Generative Adversarial Networks (GANs) and fluorescence microscopy. We applied our approach for multi-species analysis of tissue microstructure in mouse, frog, and human liver tissue, leveraging the benefits of deep learning and fluorescence microscopy. Our approach can generate accurate and high-resolution predictions of multiple tissue components, such as bile canaliculus, sinusoids and Kupffer cell shapes from images of the actin meshwork in these different species, allowing for efficient and reliable analysis. Moreover, the virtual prediction toolbox is open-source and easily accessible, making it a valuable tool for researchers with a range of expertise levels. We demonstrate the effectiveness of our methods with analysis of tissue microstructure in mouse, frog, and human liver tissue, showcasing its ability to accurately predict differences between species. Overall, our virtual prediction approach provides a powerful and efficient approach to multi-species tissue microstructure analysis. ### Competing Interest Statement The authors have declared no competing interest.