Development and validation of asthma risk prediction models using co- expression gene modules and machine learning methods

Abstract

Abstract Asthma is a chronic inflammatory disease of the airways with a strong genetic component. Because multiple genes may affect asthma, identifying differentially co-expressed genes followed by functional annotation can inform our understanding of the molecular mechanisms in asthma pathogenesis. In this study, we used airway epithelial cells (AECs) and nasal epithelial cells (NECs) datasets and implemented weighted gene co-expression network analysis (WGCNA) and machine learning (ML) techniques to develop asthma classification and predictive models. The models were validated using external bronchial epithelial cells (BECs), airway smooth muscle (ASM) and whole blood (WB) datasets. WGCNA and ML-based procedure identified 23 and 34 gene signatures that can discriminate asthmatic from control subjects in AECs (Area under the curve: AUC =0.90) and NECs (AUC = 0.99), respectively. We further validated AECs derived DEGs in BECs (AUC= 0.96), ASM (AUC= 0.72) and WB (AUC= 0.67). Similarly, NECs derived DEGs in BECs (AUC= 0.88), ASM (AUC= 0.87) and WB (AUC= 0.68). Both AECs and NECs based gene-signatures showed a strong diagnostic performance with high sensitivity and specificity. Functional annotation of NEC-derived hub genes showed several enriched pathways related to Th1 and Th2 activation pathway, while AECs-derived hub genes were significantly enriched in pulmonary fibrosis and idiopathic signaling. Several asthma related genes were prioritized including Cathepsin C (CTSC) which showed functional relevance in multiple cells relevant to asthma pathogenesis. Taken together, epithelium gene signature-based model could serve as robust surrogate model for hard-to-get tissues including BECs to improve asthma classification.