SAME: Topology-flexible transforms enable robust integration of multimodal spatial omics

Abstract

Spatial omics technologies provide complementary and layered molecular insights that span proteins, transcripts, and metabolites. However, aligning and integrating these modalities across serial tissue sections remains a computational challenge. Existing alignment methods are primarily unimodal and assume preserved topology, often failing with tissue distortions like tears, folds, or anatomical changes. Here, we present SAME (Spatial Alignment of Multimodal Expression) that introduces space-tearing transforms, a framework for controlling localized topological disruptions during cross-sectional alignment. Using integer linear programming to maximize cell type matches across the modalities, we enhance cell-type alignment accuracy by 20% compared to existing methods while preserving biologically meaningful spatial relationships. Applied to protein-RNA integration in healthy tongue tissue and lung adenocarcinoma, SAME revealed cryptic immune subpopulations that were otherwise missed by RNA-only or protein-only classification. In a separate lung adenocarcinoma study, we assayed and integrated protein and metabolomic profiles, uncovering localized mevalonic acid upregulation specifically within tumor-macrophage spatial niches and identifying targeted metabolic crosstalk invisible to single-modality approaches. SAME enables unprecedented experimental designs that leverage each modality independently while computationally recovering cross-modal spatial structure, unlocking multimodal discoveries in complex tissues.