Comparisons of Segmentation Methods for Multiplexed Spatial Proteomics

• Talha Dorez, Biological Sciences, University of Delaware

• Austin Keeler, Department of Biological Sciences, University of Delaware
• Milad Markhali, Department of Bioinformatics, University of Delaware

The nervous system is essential for organisms to respond to the world, and it is divided into the peripheral (detects the world) and central (processes information). Our research is focused on the peripheral nervous system that detects danger or harm in the world, which we can sense as pain.  Here, we are adapting a new and powerful technology, imaging mass cytometry, that can detect over 40 proteins in single-cells at the same time, to help us identify and study these neural cells that detect and sense painful cues. Importantly, this will help us understand how these cells change during pain conditions, to seek new treatments. However, imaging mass cytometry has primarily been used to understand cancer, and the analytic methods that have been developed may not work for neural cells, who have very heterogeneous and complex shapes. Our research revolves around empirically determining a method to cell segment individual neurons, which will allow downstream, high dimensional analysis of protein expression for each pain neuron. Cell segmentation is the process of labeling all the pixels where cells are located, and differentiating that with where there are no cells located and there is only background. The accuracy in this process is very crucial because the miniscule details are very sufficient for the data to be accurate, and this is necessary to study the protein expression in each cell. To determine the best method for segmentation, we have generated a large set of tissue stained sections of peripheral ganglia, manually segmented these images to generate a ground truth, and metrics of good segmentation across methods, including training our own machine learning algorithm. Once completed, we can select the method that provides the highest quality cell segmentation to add to our imaging mass cytometry workflow. We will then use this workflow to identify and explore changes in pain neurons in acute and chronic pain. This will help lead us to a conclusion and detect the different pain conditions and what treatments can be generated based off of that.