High-resolution spatially resolved proteomics of complex tissues based on microfluidics and transfer learning

Abstract

Despite recent advances in imaging- and antibody-based methods, achieving in-depth, high-resolution protein mapping across entire tissues remains a significant challenge in spatial proteomics. Here, we present parallel-flow projection and transfer learning across omics data (PLATO), an integrated framework combining microfluidics with deep learning to enable high-resolution mapping of thousands of proteins in whole tissue sections. We validated the PLATO framework by profiling the spatial proteome of the mouse cerebellum, identifying 2,564 protein groups in a single run. We then applied PLATO to rat villus and human breast cancer samples, achieving a spatial resolution of 25 μm and uncovering proteomic dynamics associated with disease states. This approach revealed spatially distinct tumor subtypes, identified key dysregulated proteins, and provided novel insights into the complexity of the tumor microenvironment. We believe that PLATO represents a transformative platform for exploring spatial proteomic regulation and its interplay with genetic and environmental factors.

Keywords: deep learning; mass spectrometry; microfluidics; spatial multi-omics; spatial proteomics; tomography.