Plasma proteomics and machine learning deliver non-invasive distinction between fibrotic hypersensitivity pneumonitis and idiopathic pulmonary fibrosis.

backgroundHypersensitivity pneumonitis (HP) manifests as fibrotic (FHP) and non-fibrotic (NFHP) phenotypes. Clinically distinguishing FHP from idiopathic pulmonary fibrosis (IPF) remains challenging owing to phenotypic overlap, despite divergent management protocols. This investigation sought to develop a plasma proteomics-based framework for differential diagnosis between these entities.

methodsA total of 119 subjects were enrolled from the Chinese Interstitial Lung Disease (ILD) National Cohort and the PORTRAY IPF Cohort between July 2018 and June 2022, comprising 32 healthy controls (HCs), 31 NFHPs, 28 FHPs, and 28 IPF patients. The plasma samples were subject to quantitative proteomic profiling, weighted gene co-expression network analysis (WGCNA), and bioinformatics analysis to identify differentially expressed proteins, core pathways, and co-expression modules. Key proteins were selected to construct and validate diagnostic models via seven machine learning algorithms.

resultsThis study delineated the plasma proteomic landscape of FHP and IPF, identifying 813 proteins. WGCNA revealed significant enrichment of the glycolysis/gluconeogenesis and pyruvate metabolism pathways, implicating metabolic reprogramming in FHP pathogenesis. Differential analysis identified nine differentially expressed proteins, from which a six-protein signature (H2BC12, SHBG, APCS, PTPRG, IGHV1-58, and GAPDH) was derived through LASSO regression and recursive feature elimination. Among seven machine learning algorithms, support vector machine (SVM) achieved the optimal performance on the independent test set with an accuracy of 71.4%, effectively discriminating FHPs from IPFs. This model represents a promising non-invasive molecular tool for the differential diagnosis of atypical interstitial lung diseases.

conclusionsThis study established the plasma proteomic landscape of FHP, linking metabolic reprogramming to disease pathogenesis and providing a machine learning framework for biomarker-guided differential diagnosis.