Comparative analysis of normalization methods for spatial transcriptomic data

Spatial transcriptomics (ST) is a promising new technology that helps to identify spatial patterns of gene expression, microenvironments, cell-cell communication networks, and contribute to our understanding of disease development. One of the first and most critical steps in data preprocessing is normalization. For similar technologies, RNA-seq and scRNA-seq, it was shown that normalization can significantly affect downstream analysis. Despite that, the role of normalization in the processing of ST data remains unexplored, and there is no consensus regarding which normalization method(s) should be used and when. Here, we explore the underlying mathematical properties of ST data and test normalization methods commonly applied to ST data - most of which were originally designed for scRNA-seq data. Specifically, we constructed a simulation platform to initially simulate single cell gene expression across two groups or conditions and, then, combine single cells into Visium-like spots with a specified density and uniformity in space. Simulated data is subsequently used to test how the density and composition of spots in combination with different normalization algorithms affects the ability to detect known differentially expressed genes. Normalization methods will also be tested on real data with known H&E annotatations. Overall, the results of this study will allow us to appraise and rank the suitability of current normalization methods as a function of tissue characteristics, and lay the groundwork for future method development.