Deconvolving Rare Cell States from Spatial Transcriptomic Data
Team: Anish Puligilla, Cody Slater, Joyce Zhou
Fig 1. Spatially visualizing the most common differentially expressed genes (DEG) in each of the three clusters of microglial cells in an LPS treated visium slice. (A) Top 4 DEGs in the activated microglial cell cluster. (B) Top 4 DEGs in the intermediate activated microglial cell cluster. (C) Top 5 DEGs in the homeostatic microglial cell cluster. Similar images for other LPS and Saline treated visium slices are available online along with our code.
Abstract:
The advent of spatial transcriptomics platforms have promoted new research endeavors seeking to understand how gene expression in cells varies with respect to the environment in the native tissue. Algorithms such as Robust Cell-Type Deconvolution (RCTD) have paved the way for scientists to deconvolve mixtures of cell types in spatial data and characterize the gene expression of specific cells within a mixture. However, algorithms such as this fail to account for the variety of states cell types can manifest as depending on other factors such as cellular environment. Microglia, due to their low tissue volume and relatively low abundance compared to other cell types in the central nervous system, have proven difficult to study. Here, to improve the characterization of microglial cell states, we use data collected from lipopolysaccharide (LPS)-injected mice to define cell state-specific clusters, integrate it with annotated clusters, and feed it into the RCTD algorithm applied to a spatial data from LPS-injected mice. Spatial mapping of microglia cell states using these extra clusters resulted in new information compared to baseline microglia detection, highlighting our approach’s potential to reveal previously masked biological insights.
My Contribution: I was responsible for visualization of microglial markers for spatial analysis and generation of baseline results for comparative analysis. I worked primarily in R using Rstudio with the Seurat package to accomplish my tasks.
Final report and code available online here