(1) Objective: Identification of potential genetic biomarkers for various glomerulonephritis (GN) subtypes and discovering the molecular mechanisms of GN. (2) Methods: four microarray datasets of GN were downloaded from Gene Expression Omnibus (GEO) database and merged to obtain the gene expression profiles of eight GN subtypes. Then, differentially expressed immune-related genes (DIRGs) were identified to explore the molecular mechanisms of GN, and single-sample gene set enrichment analysis (ssGSEA) was performed to discover the abnormal inflammation in GN. In addition, a nomogram model was generated using the R package "glmnet", and the calibration curve was plotted to evaluate the predictive power of the nomogram model. Finally, deep learning (DL) based on a multilayer perceptron (MLP) network was performed to explore the characteristic genes for GN. (3) Results: we screened out 274 common up-regulated or down-regulated DIRGs in the glomeruli and tubulointerstitium. These DIRGs are mainly involved in T-cell differentiation, the RAS signaling pathway, and the MAPK signaling pathway. ssGSEA indicates that there is a significant increase in DC (dendritic cells) and macrophages, and a significant decrease in neutrophils and NKT cells in glomeruli, while monocytes and NK cells are increased in tubulointerstitium. A nomogram model was constructed to predict GN based on 7 DIRGs, and 20 DIRGs of each subtype of GN in glomeruli and tubulointerstitium were selected as characteristic genes. (4) Conclusions: this study reveals that the DIRGs are closely related to the pathogenesis of GN and could serve as genetic biomarkers in GN. DL further identified the characteristic genes that are essential to define the pathogenesis of GN and develop targeted therapies for eight GN subtypes.
Keywords: deep learning; glomerulonephritis; immune infiltration; immune-related genes; machine learning.