Advancements and Challenges in Spatial Proteomics Imaging: From Image Handling to Downstream Analysis

Recent progress in spatial proteomics imaging has revolutionized our ability to comprehensively profile various cell types in complex tissues. High-parametric antibody panels enable the identification, quantification, and profiling of numerous cell populations and biological structures across tissues, as well as their spatial relationships. However, these advancements introduce significant technical challenges arising from imaging and the nature of the data. Issues such as out-of-focus areas in whole mounts, imaging artefacts, and persistent background fluorescence for certain markers hinder downstream semantic segmentation of biological objects. In this work, we introduce a pipeline designed to address these challenges, facilitating the extraction of meaningful biological insights from the obtained images. Additionally, we present a custom dataset comprising various types of immune cells found in the brain and brain border regions. This training dataset was used for fine-tuning semantic segmentation models, both convolution and transformer-based. Our results show that the transformer-based model has better generalization capability when trained on the same data. Finally, we highlight specific adjustments required for downstream analysis of imaging proteomics data within established single-cell transcriptomics pipelines, addressing the unique characteristics of the data.