Poster #27: Shuying Han

Quantifying Cell and Gene Repositioning Across Tissue States Using Tissue-Intrinsic Geometric Coordinates

Shuying Han, BS¹; Rong Ma, PhD¹; ¹Department of Biostatistics, Harvard T.H. Chan School of Public Health, Boston, MA, USA

Spatial omics can reveal how disease and tissue remodeling reorganize cells and molecular programs within their anatomical context. However, comparisons across conditions, timepoints, or individuals are complicated by changes in tissue geometry, because corresponding structures may no longer occupy comparable spatial coordinates after remodeling. Existing cross-section approaches often rely on spatial alignment, which becomes challenging when morphology and cellular composition differ across biological states. We developed a geometry-aware framework that enables comparison without requiring cross-section alignment. Each section is partitioned into meaningful anatomical layers, and cells within each layer are independently mapped onto an intrinsic axis from lumen to outer boundary. At the cell-type level, layer-specific positional distributions are compared using Jensen–Shannon distance and integrated across layers to identify corresponding populations and quantify spatial reorganization. For matched populations, gene-level analysis quantifies population-relative displacement, δ, as the difference between the transcript-weighted position of a gene and the mean position of its expressing cells. Section-level estimates are combined using random-effects meta-analysis with empirical null calibration. To evaluate detection of directional gene repositioning across heterogeneous tissue sections, we compared δ with a hierarchical GAM, DESpace2, and pseudobulk differential expression in simulations. δ maintained nominal Type I error under section heterogeneity, controlled the false discovery rate, remained null when the entire cell population shifted without relative gene repositioning, and achieved comparable power at matched empirical significance thresholds. Applied to a public MERFISH atlas of DSS colitis, the cell-level score ranked cognate fibroblast-to-IAF population matches among the top three candidates in 87% of section pairs, compared with 50–69% for moscot, PASTE2, CAST, and STalign. Among the matched fibroblast-to-IAF populations, 105 of 892 genes showed spatial repositioning. Mmp13 and Plau shifted toward the lumen, whereas Igfbp5 and Cxcl12 shifted toward the crypt base, suggesting spatial segregation of injury-associated proteolysis and stromal signaling during inflammation.