Fusion of Spatiotemporal and Network Models to Quantify Variations in ScRNA-seq data

Authors

• OA Egbon
• B Anchang

Abstract

A challenge with analyzing spatially resolved single-cell data collected at multiple time points is that spatial coordinates captured at different time snapshots may represent different cellular neighborhoods. Despite technological efforts to address this issue, the inherent variability among subjects hampers complete success. Naively analyzing such data using existing pipelines may be inaccurate. To mitigate these challenges, we propose SpatialDNet, an interpretable statistical model that fuses network and spatio-temporal models to quantify cellular behavior within a microenvironment. SpatialDNet takes spatial location information of cells or spots and genomic data as input. It constructs a mesh over the spatial region using a finite element method, creates network nodes from the mesh polygons, and utilizes multidimensional gene expression data to connect the nodes, forming a comprehensive network used in the spatio-temporal modeling of genes/proteins. We validated the proposed framework on a CODEX dataset obtained from the study of advanced melanoma tumors. In the experiment, CD8+ T cells activated ex-vivo with gp100 antigen and IL-2 were transferred into mice with B16-F10 tumors. Tumors were harvested and imaged at day 0, 1, 3, 5, and 12 post-treatment with 40 protein markers. We leveraged SptatialDNet to estimate the spatio-temporal variable genes, determined gene regulatory networks and annotated the trajectory of the tumor-immune cell network. We identified 25 significant proteins that are spatio-temporally varying. In addition, CD11b - Ly6G, Ly6C - Sca1, and PDL1 - H2kb significant protein interactions were identified. This work provides an interpretable framework for understanding cellular microenvironments to provide insights for personalized medicine.

Scientific Focus Area: Computational Biology

This page was last updated on Tuesday, August 6, 2024