Abstract
<title>Abstract</title> <p>Protein measurements provide critical functional information that complements transcriptomic profiles and substantially improve downstream analyses, including cell-type annotation, spatial domain identification, and tissue microenvironment characterization. Despite the growing availability of CITE-seq datasets, existing protein inference methods are primarily designed for single-cell transcriptomic data and largely ignore spatial context, limiting their ability to recover protein distributions in spatial omics datasets. Furthermore, as the number of measured proteins increases, experimental costs rise substantially, restricting the coverage of current spatial proteomics technologies. Consequently, there is an urgent need for computational frameworks that can leverage existing CITE-seq resources to both expand protein panels and infer spatially resolved protein expression, thereby extending transcriptomics-only datasets into protein-informed spatial atlases. Here, we present UNIPRO, a unified graph-based framework that learns protein-aware representations by jointly modeling transcriptomic similarity, tissue architecture, and local microenvironmental context. By adaptively integrating expression-derived and spatially derived relationships and iteratively refining graph connectivity during training, UNIPRO establishes a transferable latent space that can be applied to diverse downstream settings, including single-cell protein inference, spatial protein reconstruction, single-cell-to-spatial protein transfer, and protein-panel expansion. This unified design enables seamless integration of transcriptomics-only datasets, partially measured CITE-seq datasets, and spatial multi-omics datasets within the same computational framework. Across 16 spatial and single-cell CITE-seq datasets, UNIPRO consistently outperforms state-of-the-art methods in protein inference and imputation tasks. Moreover, the inferred protein profiles improve fine-grained cell subtype annotation and spatial domain identification, while revealing biologically meaningful tumor microenvironment markers and spatially organized functional niches. Together, UNIPRO provides a generalizable framework for extending transcriptomic measurements into protein-informed cellular and tissue atlases across diverse biological contexts.</p>