Designing Graph Neural Networks for the Analysis of Subcellular Spatial Transcriptomics

BACKGROUND 
The advent of scRNA-seq has had a profound impact on genomic and biomedical research. With the ability to profile gene expression at cellular resolution, we are able to significantly increase our understanding of cell types and cell states within different complex biological systems. At the same time, there are many questions that scRNA-seq technology cannot answer satisfactorily. Many biological processes, such as tissue development and cell signaling, are regulated in a spatially specific manner. To understand these processes, we need access to the spatial information of gene expression within tissues. Spatial transcriptomics is a groundbreaking molecular profiling method that provides single cell-resolution expression measurements with spatial resolution. The data generated from spatial transcriptomics experiments typically consists of gene expression levels associated with specific spatial coordinates. This spatially resolved transcriptomic data has been shown to improve our ability to identify cell types, infer cell-cell interactions, and study the organization of tissues and organs at a molecular level. The latest advance in high throughput single-cell transcriptomics is the emergence of spatial transcriptomics technologies with single-molecule resolution. Such subcellular spatial transcriptomics (henceforth “SST”) technologies provide not only the transcript abundance of each gene in a cell, but they also provide the precise subcellular locations of those transcripts, thus materializing a true spatial map of the transcriptome. Figure 1: Visualization of the graph generated for a randomly sampled cell. The nodes are colored based on the gene. 

Our lab has been working on tools for the analysis of SST (subcellular spatial transcriptomics) data and has developed the “Intracellular Spatial Transcriptomic Analysis Toolkit” (InSTAnT). This work has been described in a manuscript that is currently in revision, and I have contributed to this manuscript. InSTAnT is a toolkit that detects gene pairs and modules that co-localize within cells using specialized statistical tests. InSTAnT was used to discover several novel cell type-specific gene pair co-localizations in the brain, and its promising results demonstrate potential for future research.