Gene-expression biomarkers of lung allograft rejection identified through transcriptomic analysis and machine learning

Lung allograft rejection remains a major cause of morbidity and mortality after lung transplantation. Histopathological assessment of transbronchial biopsies is the gold standard diagnostic method but is limited by sampling error and inter-observer variability. We aimed to identify reproducible candidate gene-expression biomarkers associated with rejection using transcriptomic analysis and machine-learning approaches. Microarray data from 457 transbronchial biopsies (GSE150156) were analyzed to identify differentially expressed genes. Functional enrichment and protein-protein interaction analyses were performed. Candidate biomarkers were prioritized using L1-regularized logistic regression and random forest algorithms and externally validated in a dataset of 190 third-bronchial-bifurcation mucosal biopsies (GSE125004).We identified 1,412 differentially expressed genes between rejection/injury-associated and non-rejection biopsies. Machine-learning analysis prioritized 18 candidate biomarkers with area under the receiver operating characteristic curve exceeding 90%. External evaluation identified 13 significantly differentially expressed genes, of which 11 showed concordant directions of expression across both datasets. Among these, CYBB and LCP2 emerged as candidate biomarkers not previously evaluated as individual biomarkers in lung transplant rejection. Integrative transcriptomic and machine-learning analyses identified reproducible candidate biomarkers for molecular assessment of lung allograft rejection. Prospective studies are warranted to validate these findings and support development of targeted diagnostic assays.