scVIVA - A Probabilistic Framework for Representation of Cells and Their Environments

Learn about scVIVA, a deep generative model that advances spatial transcriptomics analysis by incorporating both cell-intrinsic and neighboring gene expression profiles to create comprehensive cellular embeddings. Discover how this probabilistic framework addresses limitations of current methods that only consider individual cell gene expression by integrating environmental context from surrounding cells. Explore the model's ability to produce stochastic embeddings of cell states and normalized gene expression profiles, enabling fine-grained cell partitioning that reflects both internal cellular states and surrounding tissue characteristics. Understand how scVIVA's generative capabilities facilitate hypothesis testing for differential expression between tissue niches, demonstrated through the identification of spatially-restricted tumor-promoting endothelial populations in breast cancer and niche-associated T cell states shared across multiple cancer types. Gain insights into the implementation and availability of this open-source tool within the scvi-tools.org ecosystem for spatial transcriptomics research.