Microenvironment-aware transcriptome reconstruction in spatial transcriptomics

Abstract Imaging-based spatial transcriptomics offers single-cell resolution but measures limited panels dominated by identity-defining genes, leaving transcriptome-wide variation unobserved. Existing approaches for predicting unmeasured genes rely mainly on shared-gene alignment, which recovers identity-related expression but fails to capture subtle microenvironment-driven variation within a cell type. We introduce Emerge, a framework that reconstructs transcriptome-scale expression by jointly modeling intrinsic transcriptional manifolds from single-cell RNA sequencing and the extrinsic niche organization observed in spatial data within a type-constrained optimal transport formulation. Across fourteen MERFISH and Xenium datasets from neural and tumor tissues, Emerge improves prediction accuracy, spatial coherence and recovery of within-type heterogeneity. The reconstructed transcriptomes reveal microenvironment-stratified astrocyte, stromal and fibroblast states that are only partially captured by measured panels or existing prediction approaches, establishing Emerge as a generalizable foundation for context-aware reconstruction in spatial biology.