Identification of AKI signatures and classification patterns in ccRCC based on machine learning

BACKGROUND: Acute kidney injury can be mitigated if detected early. There are limited biomarkers for predicting acute kidney injury (AKI). In this study, we used public databases with machine learning algorithms to identify novel biomarkers to predict AKI. In addition, the interaction between AKI and clear cell renal cell carcinoma (ccRCC) remain elusive. METHODS: Four public AKI datasets (GSE126805, GSE139061, GSE30718, and GSE90861) treated as discovery datasets and one (GSE43974) treated as a validation dataset were downloaded from the Gene Expression Omnibus (GEO) database. Differentially expressed genes (DEGs) between AKI and normal kidney tissues were identified using the R package limma. Four machine learning algorithms were used to identify the novel AKI biomarkers. The correlations between the seven biomarkers and immune cells or their components were calculated using the R package ggcor. Furthermore, two distinct ccRCC subtypes with different prognoses and immune characteristics were identified and verified using seven novel biomarkers. RESULTS: Seven robust AKI signatures were identified using the four machine learning methods. The immune infiltration analysis revealed that the numbers of activated CD4 T cells, CD56(dim) natural killer cells, eosinophils, mast cells, memory B cells, natural killer T cells, neutrophils, T follicular helper cells, and type 1 T helper cells were significantly higher in the AKI cluster. The nomogram for prediction of AKI risk demonstrated satisfactory discrimination with an Area Under the Curve (AUC) of 0.919 in the training set and 0.945 in the testing set. In addition, the calibration plot demonstrated few errors between the predicted and actual values. In a separate analysis, the immune components and cellular differences between the two ccRCC subtypes based on their AKI signatures were compared. Patients in the CS1 had better overall survival, progression-free survival, drug sensitivity, and survival probability. CONCLUSION: Our study identified seven distinct AKI-related biomarkers based on four machine learning methods and proposed a nomogram for stratified AKI risk prediction. We also confirmed that AKI signatures were valuable for predicting ccRCC prognosis. The current work not only sheds light on the early prediction of AKI, but also provides new insights into the correlation between AKI and ccRCC.