SPINE: Context-guided iterative protein inference from spatial transcriptomes

doi:10.1016/j.eswa.2026.133652

SPINE: Context-guided iterative protein inference from spatial transcriptomes

2026-07-01

0

PRE

AI

DOI：10.1016/j.eswa.2026.133652

Original

Original request for help

Save

Abstract

En 中文

Spatially resolved protein abundance is critical for understanding cellular function, yet remains difficult to measure at scale, motivating its inference from transcriptomic data as an important intelligent prediction problem in spatial multi-omics. However, existing methods are primarily developed in the single-cell setting and formulate this task as independent point-wise prediction, ignoring the structured spatial dependencies induced by local cellular interactions and tissue architecture. Here, we present SPINE, a neighborhood-guided, flow-matching-inspired iterative refinement framework for protein inference from spatial transcriptomes. Rather than treating protein prediction as independent point-wise regression, SPINE reformulates this task as a conditional refinement problem, in which an initial protein prior is iteratively updated toward the target protein profile under transcriptomic and spatial-neighborhood guidance. To incorporate tissue-context information, SPINE combines neighborhood-aware spatial modeling with expression-based graph structure, while an auxiliary reconstruction branch helps preserve transcriptomic semantics during cross-modal inference. Across paired spatial RNA-protein datasets, SPINE achieves superior overall prediction performance compared with representative RNA-to-protein and multimodal baselines, and better preserves protein-informed biological structure in downstream spatial analyses.

Keywords:

Spatial transcriptomics

Protein inference

Spatial multi-omics

Iterative refinement

Modality completion

SPINE: Context-guided iterative protein inference from spatial transcriptomes

Abstract

Journal

Organization

Cited Papers