CellFuse Enables Multi-modal Integration of Single-cell and Spatial Proteomics Data

Abstract

Single-cell and spatial proteomic technologies capture complementary biological information, yet no single platform can measure all modalities within the same cell. Most existing integration methods are optimized for transcriptomic data and rely on a large set of shared, strongly linked features, an assumption that often fails for low-dimensional proteomic modalities. We present CellFuse, a deep learning-based, modality-agnostic integration framework designed specifically for settings with limited feature overlap. CellFuse leverages supervised contrastive learning to learn a shared embedding space, enabling accurate cell type prediction and seamless integration across modalities and experimental conditions. Across a range of datasets including healthy PBMCs, bone marrow, CAR-T-treated lymphoma, and healthy and tumor tissues--CellFuse consistently outperforms existing methods in both integration quality and runtime efficiency. It maintains high accuracy even in the presence of missing markers and rare cell types, and performs robustly in cross-dataset comparisons, making it a powerful tool for scalable and high-fidelity single-cell data integration in basic and translational research.