Computational approaches for the analyses of spatial profiling technologies

点击翻译按钮获取中文摘要

Spatial transcriptomics (ST) technology has enabled geographical gene expression profiling in tumors. However, each tumor ST spot contains diverse immune and malignant cells, with cell densities that vary significantly across tissue regions. Therefore, the cell type decomposition in tumor ST data remains a challenge that existing methods for bulk tumors and general ST data cannot resolve. We developed Spatial Cellular Estimator (SpaCE) to infer cell identities and intercellular interactions from tumor ST data. SpaCE first estimates cancer cell clonal abundance through modeling segmental copy number variations. A constrained regression model then calibrates local cell densities and determines immune and stromal cell lineage fractions. SpaCE provides higher accuracy than existing methods based on simulated samples of known composition and human tumors with double-blind pathology annotations on hematoxylin and eosin images. Moreover, coupling cell fraction results with ligand-receptor spatial expression, SpaCE reveals how intercellular interactions at the tumor-immune interface promote cancer progression.