A1085
Title: Spatial Poisson-lognormal pathway model for detecting spatially expressed (SE) genes in spatial transcriptomics
Authors: Emmanuel Sarfo Fosu - Baylor University (United States) [presenting]
Joon Jin Song - Baylor University (United States)
Thierry Chekouo - University of Minnesota (United States)
Abstract: Spatial transcriptomics (ST) enables high-resolution mapping of gene expression across tissues, offering spatial insights into cellular organization, tissue development, disease progression, and treatment response. One key objective in ST analysis is to identify spatially expressed (SE) genes. Most existing methods, however, ignore gene-gene dependencies. The aim is to propose a spatial Poisson lognormal model that uses biological pathways to jointly capture both spatial and gene-level dependencies. Given the high dimensionality of ST data, a non-spatial conditional autoregressive (CAR) prior is adopted that models gene dependencies by borrowing external biological knowledge. The Bayesian model simultaneously detects gene clusters of non-SE and SE. By integrating these localized gene dependencies into a hierarchical spatial framework, the model improves sensitivity and interpretability in detecting SE genes. Simulation studies and applications to real ST datasets demonstrate enhanced power and accuracy compared to existing univariate methods, while leveraging biologically meaningful gene-gene relationships.
