Guo S, Wang E, Wang B, Xue Y, Kuang Y, Liu H. Comprehensive Multiomics Analyses Establish the Optimal Prognostic Model for Resectable Gastric Cancer : Prognosis Prediction for Resectable GC.
Ann Surg Oncol 2024;
31:2078-2089. [PMID:
37996637 DOI:
10.1245/s10434-023-14249-x]
[Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Key Words] [MESH Headings] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 06/12/2023] [Accepted: 08/14/2023] [Indexed: 11/25/2023]
Abstract
BACKGROUND
Prognostic models based on multiomics data may provide better predictive capability than those established at the single-omics level. Here we aimed to establish a prognostic model for resectable gastric cancer (GC) with multiomics information involving mutational, copy number, transcriptional, methylation, and clinicopathological alterations.
PATIENTS AND METHODS
The mutational, copy number, transcriptional, methylation data of 268, 265, 226, and 252 patients with stages I-III GC were downloaded from the TCGA database, respectively. Alterations from all omics were characterized, and prognostic models were established at the individual omics level and optimized at the multiomics level. All models were validated with a cohort of 99 patients with stages I-III GC.
RESULTS
TTN, TP53, and MUC16 were among the genes with the highest mutational frequency, while UBR5, ZFHX4, PREX2, and ARID1A exhibited the most prominent copy number variations (CNVs). Upregulated COL10A1, CST1, and HOXC10 and downregulated GAST represented the biggest transcriptional alterations. Aberrant methylation of some well-known genes was revealed, including CLDN18, NDRG4, and SDC2. Many alterations were found to predict the patient prognosis by univariate analysis, while four mutant genes, two CNVs, five transcriptionally altered genes, and seven aberrantly methylated genes were identified as independent risk factors in multivariate analysis. Prognostic models at the single-omics level were established with these alterations, and optimized combination of selected alterations with clinicopathological factors was used to establish a final multiomics model. All single-omics models and the final multiomics model were validated by an independent cohort. The optimal area under the curve (AUC) was 0.73, 0.71, 0.71, and 0.85 for mutational, CNV, transcriptional, and methylation models, respectively. The final multiomics model significantly increased the AUC to 0.92 (P < 0.05).
CONCLUSIONS
Multiomics model exhibited significantly better capability in predicting the prognosis of resectable GC than single-omics models.
Collapse