Identification and validation of a novel prognosis prediction model in adrenocortical carcinoma by integrative bioinformatics analysis and statistics and machine learning

Abstract

Background: Adrenocortical carcinoma (ACC) is a rare malignancy with poor prognosis. Thus, we aimed to establish a gene signature to predict the prognosis for ACC. Methods: Firstly, “WGCNA” package was used to construct a co-expression network and screen key module. Secondly, survival associated genes were identified by performing survival analysis. Thirdly, regression models were constructed by using the Ridge, ELASTIC-NET, and LASSO methods. Time-dependent ROC analysis, Cox regression analysis, GSEA, DCA and nomogram were performed to validate the model. Fourthly, mutations and CNVs of genes in the model were explored. Finally, LDA, KNN, SVM, PPI network and meta-analysis were used screened and validated meaningful prognostic biomarkers.Results: Two key modules were selected and 93 survival associated genes were identified. Furthermore, 11 models were constructed and two models were further selected, which were validated in each dataset (training set, internal validation set, GSE19750, and GSE76021). Model 2 was further identified as the best model (training set: survival analysis: p < 0.0001; AUC: 0.92 at 1 year, 0.91 at 3 years and 0.95 at 5 years). In genes in the best model, MKI67 was altered most (12%). Six hub genes were further analyzed by constructing a PPI network and validated by meta-analysis. Conclusion: In summary, we constructed and validated a prognostic multi-gene model and six powerful prognostic biomarkers, which might be useful instruments for predicting the prognosis of ACC patients.