SM2ST

Overview

Spatial transcriptomics, leveraging in situ sequencing or in situ hybridization, reveals the spatial distribution of gene expression at single-cell or sub-cellular resolution within tissue sections. Concurrently, spatial metabolomics employs mass-spectrometry imaging (MSI), specifically MALDI-MS, to map metabolite distributions and provide molecular insight into metabolic activity. Together, these techniques furnish complementary information on gene expression and metabolic dynamics for interrogating the tissue microenvironment. However, the drift of MSI during the instrument’s long-term acquisition and the heterogeneity in its spatial chemical matrix coverage introduce noise that complicates cross-modal spatial alignment. Here we introduce an innovative multimodal registration framework that adopts hematoxylin-and-eosin (H&E) stained image as a bridging modality to compute adaptive affine transformations, aligning MSI with histological references and achieving landmark registration errors below 10 pixels. This process enables spatial transcriptomics and spatial metabolomics to be projected into a unified coordinate system. Building upon this, we propose a novel architectural framework that integrates generative adversarial networks (GANs) with autoencoders. This innovative approach enables effective denoising of metabolic ion signals and remapping onto spatial transcriptomic loci, thereby achieving point-to-point co-registration between the two modalities. The proposed method effectively addresses spatial heterogeneity across diverse data types, thereby enabling the development of a unified spatial multi-omics analytical framework. Furthermore, via a self-supervised super-resolution model termed STMGraph, we compensate for the limited sensitivity of high-mass molecules in conventional high-resolution MALDI-MS, effectively enhancing spatial metabolomic resolution.