Multimodal AI-enabled mass spectrometry-based expansion proteomics for whole-slide at single-cell resolution

Abstract

Deep, quantitative proteome coverage at single-cell resolution across entire tissue sections remains a major challenge for mass spectrometry-based spatial proteomics. Here, we introduce an AI-empowered filter-aided expansion proteomics (FAXP) framework that combines FAXP with convolutional neural network (CNN)-based spatial inference to achieve whole-slide, single-cell-resolved proteomics. By combining tissue expansion with orthogonal laser capture microdissection to obtain whole-slide linear strip-resolved proteome measurements, we develop HetuNet, a CNN-based model that integrates these sparse orthogonal data with high-dimensional imaging-derived contexts to reconstruct comprehensive two-dimensional spatial protein expression landscapes. In mouse liver, this approach captures zone-specific protein patterns and continuous pathway-level gradients, validated by spatial transcriptomics. In colorectal cancer tissues, it resolves proteome-defined epithelial states, revealing functional divergence via differential epithelial-mesenchymal transition and necroptosis activation. Together, this framework enables deep, scalable spatial proteomic mapping across whole tissues at single-cell resolution, unlocking previously inaccessible insights into tissue organization and function.