Identification and Validation of a Novel Prognosis Prediction Model in Adrenocortical Carcinoma by Integrative Bioinformatics Analysis, Statistics, and Machine Learning

Abstract

Adrenocortical carcinoma (ACC) is a rare malignancy with poor prognosis. Thus, we aimed to establish a potential gene model for prognosis prediction of patients with ACC. First, weighted gene co-expression network (WGCNA) was constructed to screen two key modules (blue: P = 5e-05, R^2 = 0.65; red: P = 4e-06, R^2 = −0.71). Second, 93 survival-associated genes were identified. Third, 11 potential prognosis models were constructed, and two models were further selected. Survival analysis, receiver operating characteristic curve (ROC), Cox regression analysis, and calibrate curve were performed to identify the best model with great prognostic value. Model 2 was further identified as the best model [training set: P < 0.0001; the area under curve (AUC) value was higher than in any other models showed]. We further explored the prognostic values of genes in the best model by analyzing their mutations and copy number variations (CNVs) and found that MKI67 altered the most (12%). CNVs of the 14 genes could significantly affect the relative mRNA expression levels and were associated with survival of ACC patients. Three independent analyses indicated that all the 14 genes were significantly associated with the prognosis of patients with ACC. Six hub genes were further analyzed by constructing a PPI network and validated by AUC and concordance index (C-index) calculation. In summary, we constructed and validated a prognostic multi-gene model and found six prognostic biomarkers, which may be useful for predicting the prognosis of ACC patients.