Project Summary/Abstract Understanding the spatial landscape of gene expression in tissues is a fundamental question for human health and disease. Applications range from identifying the spatial organization of cell types to dysregulation of spatial-dependent gene expression associated with disease. Advances in technologies, such as spatially-resolved transcriptomics (SRT), provide a wealth of data to investigate these questions. Furthermore, SRT combined with advances in long-read RNA-sequencing enable applications such as identifying spatial-dependent splicing variation and allele specificity in healthy and disease states, such as cancer or neurodegenerative disorders. Recent SRT studies are generating datasets across multiple samples (different donors or adjacent tissue sections), but researchers analyze samples independently because there lack computational tools for datasets with multiple samples. In contrast, when samples are jointly analyzed together, the statistical power is increased to detect differences with greater accuracy and precision. The lack of tools to analyze SRT data with multiple samples is a significant knowledge gap that limits are ability to refine the molecular causes and consequences of diseases that can be targeted for prevention and treatment. My research program develops scalable computational methods and open-source software for biomedical data analysis, in particular single-cell and spatial transcriptomics data, leading to an improved understanding of human health and disease. Here, our goal is to focus on developing scalable computational methods and software for data from spatial and long-read technologies with multiple samples and experimental conditions to accurately (1) predict spatial domains of tissues across multiple samples, (2) identify differences in spatial gene expression across experimental conditions or biological groups with multiple samples in each group, and (3) identify differential splicing variation across spatial domains or experimental conditions. The rationale for the proposed work is that the computational tools developed will enable substantial advances in our understanding of the spatial landscape of gene expression on distinct scales from cells to tissues to individuals. The significance of this proposal is substantial with broad impact for researchers increasingly using these imaging and genomic data, such as large-scale consortia generating spatial atlases across multiple samples, but also the proposed methods will be relevant to a wide variety of scientific disciplines that leverage high-dimensional data in a spatial context, such as environmental and mobile health. The project builds on my past experience in developing computational methods and open-source software for scalable clustering and identifying differences in gene expression at the single-cell level. The creation of well-documented, open-source software expands the impact of this work to other researchers aiming to understand the spatial landscape of gene expression in a variety of disease settings.

Project Narrative High-throughput spatial technologies are poised to become ubiquitous and transformative in precision medicine, including the diagnosis, prognosis, and therapeutic response in disease. However, researchers who aim to appropriately interpret these data face new challenges with increasingly complex data and experimental designs, such how to harmonize and jointly analyze multiple spatial samples simultaneously with images and omics data, or how to identify spatial-dependent difference across experimental groups, such as healthy and disease states, each group with multiple donors or adjacent tissue sections. Our long-term goal is to address these questions by developing statistically-rigorous computational tools for spatial and long-read technologies with increased statistical power to detect spatial domains or differences in spatial-dependent gene expression in large atlas-scale spatially-resolved transcriptomics data.