SPOmiAlign: A modality-agnostic framework for robust and scalable spatial multimodal alignment via feature matching

Abstract

Multimodal spatial omics enables systematic characterization of tissue organization by jointly profiling molecular features across transcriptomic, proteomic, and metabolomic layers. However, integrative analysis across sections and modalities is frequently hindered by non-linear tissue distortions, mismatched spatial resolutions, and the absence of shared molecular features. To address these challenges, we introduce SPOmiAlign, a modality-agnostic framework based on uncertainty-aware dense feature matching for spatial multimodal alignment. The framework achieves second-level runtime efficiency and enables accurate alignment across heterogeneous spatial omics datasets, as well as between spatial omics and histology, without manual effort or modality-specific tuning. Across diverse datasets, SPOmiAlign consistently achieves higher alignment accuracy than existing state-of-the-art methods. We further demonstrate its utility through automated registration of spatial omics data to a common coordinate framework (CCF), enabling anatomical annotation. Applying SPOmiAlign to spatial multi-omic data demonstrates how accurate multimodal alignment is essential for integrative spatial multi-omic analysis, enabling the identification of coherent spatial domains across modalities and supporting a broad range of downstream spatial analysis.