Decoding the Tumor Microenvironment from Spatial Analysis of Multiplexed Tissue Imaging Data

Multiplexed tissue imaging platforms enable the study of tumor immune microenvironment at spatially resolved single cell resolution. A unique challenge for whole-slide multiplexed tissue staining lies in the resulting megapixel images that require innovative and scalable analytical solutions for real world application. In this talk, I will introduce statistical and computational methods and tools developed in my group to facilitate large-scale multiplexed immunoflorescence (mIF) tissue image analysis. First, I will present a deep learning tool called CellGate for cell type classification in whole-slide mIF images. CellGate was trained on millions of labeled single-cell images, learning the variation in marker staining patterns in line with where the marker is expressed and localized (e.g., nucleus, cytoplasm, membrane), and different patterns of noise and background staining inherent to the biology and chemistry. I will also introduce SpatialTopic, a method to identify spatial signatures based on a niche analysis of cell type composition in cellular neighborhoods. SpaTopic adapts a Bayesian topic model originally developed for computer vision to spatial single cell data to identify spatial immune signatures (e.g., tertiary lymphoid like structure) in the tumor microenvironment.