Construction of Predictive Model of Interstitial Fibrosis and Tubular Atrophy (IFTA) After Kidney Transplantation with Machine Learning Algorithm

Abstract

Abstract Background: Interstitial fibrosis and tubular atrophy (IFTA) are the histopathological manifestations of CKD and one of the causes of long-term renal loss in transplanted kidneys. The purpose of our study is to screen IFTA-related genes with higher importance scores through Random Forest (RF) and further construct IFTA diagnostic model through Artificial Neural Networks (ANNs). Methods: We screened all 162 “kidney transplant” related cohorts in the GEO database and obtained 5 data sets (training sets: GSE98320 validation sets: GSE22459, GSE53605 and GSE76882 survival sets: GSE21374). Differentially expressed genes (DEGs) analysis, Gene Ontology (GO), Kyoto Encyclopedia of Genes and Genomes (KEGG), Random Forest (RF), Artificial Neural Network (ANN), Unsupervised Clustering analysis, CIBERSORT analysis were used to analyze the data. Results: A total of 108 common DEGs were identified by taking the intersection of the DEGs of our training sets and validation sets. A total of 15 top IFTA-specific DEGs were screened through the RF, then was used to build ANNs models. The model has good performance in both the training sets [GSE98320 (AUC = 0.9560)] and the validation sets [GSE22459 (AUC = 0.720), GSE53605 (AUC =0.938), GSE76882 (AUC = 0.781)], indicating that we have avoided overfitting while improving the accuracy. Furthermore, samples of survival sets are divided into two clusters using consensus clustering algorithm basing on the expression of 15 top IFTA-specific DEGs. We found significant differences between the two subgroups by survival analysis, and further enrichment analysis and immune cell infiltration analysis were conducted to further explore the causes of survival differences. Conclusion: we identified key biomarkers of IFTA and developed a new IFTA classification model, basing on the combination of RF and ANNs.