Background: The unknown etiology of sarcoidosis with variable clinical features leads to delayed diagnosis and limited therapeutic strategies. Hence, exploring the latent mechanisms and constructing an accessible and reliable diagnostic model of sarcoidosis is vital for innovative therapeutic approaches to improve prognosis.
Methods: This retrospective study analyzed transcriptomes from 11 independent sarcoidosis cohorts, comprising 313 patients and 400 healthy controls. The weighted gene co-expression network analysis (WGCNA) and differentially expressed gene (DEG) analysis were performed to identify molecular biomarkers. Machine learning was employed to fit a diagnostic model. The potential pathogenesis and immune landscape were detected by bioinformatics tools.
Results: A 10-gene signature SARDS consisting of GBP1, LEF1, IFIT3, LRRN3, IFI44, LHFPL2, RTP4, CD27, EPHX2, and CXCL10 was further constructed in the training cohorts by the LASSO algorithm, which performed well in the four independent cohorts with the splendid AUCs ranging from 0.938 to 1.000. The findings were validated in seven independent publicly available gene expression datasets retrieved from whole blood, PBMC, alveolar lavage fluid cells, and lung tissue samples from patients with outstanding AUCs ranging from 0.728 to 0.972. Transcriptional signatures associated with sarcoidosis revealed a potential role of immune response in the development of the disease through bioinformatics analysis.
Conclusions: Our study identified and validated molecular biomarkers for the diagnosis of sarcoidosis and constructed the diagnostic model SARDS to improve the accuracy of early diagnosis of the disease.
Keywords: WGCNA; biomarker; diagnostic model; functional analysis; immune infiltration; machine learning; sarcoidosis.
Copyright © 2022 Duo, Liu, Li, Wang, Zhang, Weng, Zheng, Fan, Wu, Xu, Ren and Cheng.