Computational methods for in situ spatial transcriptomics

Project Summary/Abstract The explosive growth of spatial transcriptomics technologies has revolutionized the study of tissue spatial architecture and development. In contrast to microdissection and spatial barcoding methods, which may not always achieve single-cell resolution, in situ spatial transcriptomics provides unparalleled detail by

recording the spatial locations of individual RNA transcripts. Though specialized computational methods have been developed to tackle the unique challenges of analyzing in situ spatial transcriptomics data, substantial obstacles still exist in accurately identifying cell boundaries, distinguishing cell clusters

and cell types, and understanding cells' interactions with their microenvironment. In this project, we propose to develop a suite of computational tools to address these challenges. First, we will develop a generally applicable framework for optimized cell segmentation that integrates RNA spatial location

information with imaging information, capitalizing on the latest segmentation algorithms, such as those based on transformers. To evaluate their performances, we will generate a benchmarking dataset by manually annotating cell boundaries in real in situ data from various tissues and disease conditions.

Second, we will establish a framework of cell clustering and cell type annotations for in situ data, blending gene expression information with cell morphology and cell density information learned from images. We will also test different combinations of computational methods for data transformation, dimension reduction, and clustering. The performance will be evaluated on simulated datasets derived from

single-cell RNA-seq data with ground truth cell clusters and cell types. Finally, we will develop a flexible method to systematically study cellular microenvironment for data from both in situ and other types of spatial profiling technologies, taking into account diverse cell types, molecules, and different types of cell-cell

interactions based on spatial proximity. These methods will facilitate deeper understandings of the spatial distributions and interactions of different cell types, providing new biological insights into cell senescence, tumor microenvironment, and more.