SATOMI

PhD biology students in the past would have to make the best of their monotonous lab tasks: “We would hold regular cell-counting parties,” relates Prof. Dr. Dietrich Kohlheyer of Forschungszentrum Jülich. “We had beer and pizza, and at recurring intervals we would look through the microscope to count how the bacteria were multiplying over time.” However, when it comes to analysing more complex cell structures, counting is no longer an option.

This is where the research project comes in, which is aimed at developing a deep-learning method for analysing images from microscopes. “We will then use this method to watch bacteria as they grow,” Kohlheyer says. As the bacterial cultures grow under the microscope, pictures will be taken at set intervals. The task goes beyond simple counting: “We also want to know how the individual cells behave. So, for example, what daughter cells came about from the division of a parent cell?” says Dr. Katharina Nöh. She and her workgroup at FZ Jülich are developing software that can be used to track a cell family tree of sorts from images chock full of bacteria. The method of choice is called “probabilistic multi-object-tracking”, and the data processing for this method is being developed together with Prof. Ralf Mikut of KIT and Dr. Hanno Scharr of FZ Jülich. It is intended to work for all kinds of morphologies. Different bacteria namely develop different structures as they grow: some have daughter cells that split off, some have vesicles that burst open, while others grow out long like the branch of a tree. The aim is to develop artificial intelligence that can follow all of these variants, and even distinguish between individual cells among several bacterial species living and growing together.