Host: European Molecular Biology Laboratory (EMBL), DE.

PhD Awarding institution: Heidelberg University (UHEI), DE.

Primary Supervisor: Dr. Anna Kreshuk.

Project duration: 36 months.

Project description: Morphological analysis of cells in tissues is contingent on our ability to perform cell segmentation in vEM image stacks. While an expert biologist can delineate cell membranes in vEM data manually, this approach does not scale beyond a few cells of interest, making automated cell segmentation algorithms a prerequisite for whole-organism and cross-organism analysis. State-of-the-art segmentation methods, such as the ones used in recent atlas-building efforts or large-scale connectomics studies, operate in the paradigm of supervised machine learning, training the neural networks on densely (manually) annotated reference sub-volumes of the data. The first objective of this project is to significantly relax the requirements to the amount and density of training data, enabling practical use of such methods for multiple datasets from different species acquired on different microscopes. The second objective is to establish a set of unbiased descriptors of the segmented cells, including both shape and ultrastructure features which will allow to group and explore morphologically similar cells across organisms. In addition to cellular-level descriptors, we will develop methods to characterise areas of tissue and enable the search for similar features across datasets. In more detail, we will approach automatic segmentation as both a transfer learning problem from the few available datasets, including the Platynereis dumerilii atlas, and a weakly supervised segmentation problem where manual labels for novel ultrastructural elements will be introduced in a targeted manner. Building on recent success stories for single-cell morphology analysis, our cell descriptors will be based on unsupervised machine learning, exploiting biological priors and targeted manual labels for validation and quality control.

Application deadline: 11th March 2024

How to apply: Candidates shall apply via the EMBL International PhD program at the following link and express their interest in this position within the group of Anna Kreshuk.