ReSPIRE: A Novel Regulatory Spatial Protein Activity Model to Uncover the IPF-LUAD Transition

Overview

The molecular mechanisms driving the transition from Idiopathic Pulmonary Fibrosis (IPF) to Lung Adenocarcinoma (LUAD) remain unresolved. Since 70% of IPF-associated tumors arise within existing fibrotic tissue, characterizing this transition demands spatially resolved molecular profiling at single-cell resolution, a capability existing methods cannot provide without sacrificing either spatial context or molecular depth. ReSPIRE addresses this by predicting the activity of more than 7000 regulatory proteins at single-cell spatial resolution directly from histopathology, combining frozen UNI pathology foundation model embeddings with a residual multilayer perceptron to generate approximately 3.5 billion spatially resolved protein activity values per tissue without sequencing. Protein activity, quantified as Normalized Enrichment Scores through regulon-based inference, captures functional signaling states invisible to transcript-level analysis, enabling spatially resolved identification of diagnostic and prognostic biomarkers with direct therapeutic relevance. Trained across 2.5 million Visium HD spots from six human lung tissues under leave-one-out cross-validation, ReSPIRE achieves Pearson r = 0.6083, RMSE = 0.0935, MAE = 0.0653, and 76% directional concordance against proteins with established roles in the IPF-LUAD transition. Spatial co-activation analysis identifies a conserved regulatory architecture between disease states (r = 0.860). Furthermore, the study identifies STEAP1, S100A10, S100A11, and ITGA6 as a novel signature of bridge proteins that serve as potential therapeutic targets in the pathological transition from IPF to LUAD. These findings establish ReSPIRE as a transformative, cost-effective framework for spatially resolved protein activity inference, opening a direct path from routine clinical histopathology to molecular discoveries across cancer and chronic disease at the clinical scale.