Abstract
Spatial transcriptomics increasingly generates multi-slice atlases spanning developmental, regenerative, and disease-associated states. Identifying genes with spatial pattern remodeling across these states is essential for understanding tissue reorganization and the underlying gene programs, yet few existing methods systematically detect such changes while accounting for variation among replicate slices. Here we introduce DeSPA, a statistical framework that models gene-specific spatial variation within each slice using a linear mixed model, tests for cross-state remodeling by aggregating evidence across replicate slices, and organizes identified genes into interpretable spatial co-expression modules that reveal coordinated spatial programs. Simulations show that DeSPA achieves high power while controlling the false discovery rate under inter-slice heterogeneity. In mouse embryogenesis, it recovers programs linked to extracellular matrix organization, muscle formation, neural development, and erythrocyte development. In diabetic testis, it reveals disruption of a late-spermatid spatial program consistent with known spermatogenic dysfunction, while maintaining stable gene rankings under coordinate rotation. In axolotl brain, regeneration preserves a developmental-like biosynthetic backbone but reorganizes it into regeneration-specific modules. Across two mouse intracerebral hemorrhage models, DeSPA resolves a conserved spatial injury program alongside model-specific synaptic and tissue-response programs. Together, these findings establish DeSPA as a scalable and interpretable framework for detecting spatial pattern remodeling and resolving coordinated spatial programs across diverse biological states.