Towards a Thinking X-ray Microscope: Deep Learning to Predict Fluorescence Labels of Cellular Organelles in SXT

Despite the recent progress in imaging cells by various techniques, there is a fundamental lack of connecting molecular identity with cellular ultrastructure. Fluorescence labeling and sensitive imaging can visualize the molecular components underlying the machinery of living cells but does not resolve the structural context in which cellular processes take place. Soft X-ray tomography (SXT) is a revolutionizing technique to study the ultrastructure of cells in a hydrated state with an isotropic resolution of about 50 nm in all three dimensions. We will apply advanced deep learning methods to transfer knowledge between both imaging modalities and thereby to map molecular identity to cellular ultrastructure. This will revolutionize correlative imaging and add a new dimension to well-established fluorescence microscopy applications. At the same time, our results will make X-ray microscopy broadly applicable to life scientists in academia and industry.