Original Article
Risk stratification for prostate cancer via the integration of omics data of The Cancer Genome Atlas
Abstract
Background: Prostate carcinoma (PCa) is the second most common malignant disease in men. Despite evidence that prostate-specific antigen (PSA) screening can reduce PCa specific metastasis and death, it has not accepted by various health authorities. In fact, a broad range of heterogeneity causes different clinical and molecular behavior of PCa; risk stratification thus is helpful in guiding the optimal treatment of PCa patients.
Methods: Here, we proposed a novel frame to perform risk stratification and identify PCa risk-related genes via integration of PCa omics data from The Cancer Genome Atlas (TCGA). Firstly, risk genes were extracted by applying Cox regression model. Secondly, consensus non-negative matrix factorization (CNMF) cluster algorithm was applied to RNA-seq expression data of these risk genes, and PCa patients were divided into two subtypes (named as low risk and high risk respectively). Thirdly, by combining with survival analysis and differential expression analysis based on two subtypes, a PCa subtype-related network module was identified.
Results: The identified network module can serve as biomarkers such as SRC to predict PCa risk. In particular, we observed the obvious differences in the DNA methylation profile and copy number variation (CNV) of genes involved in the module between two PCa subtypes.
Conclusions: The framework proposed in this paper provides an effective strategy for the comprehensive analysis of TCGA omics data and can help highlight the prevention and treatment stratification for PCa patients.
Methods: Here, we proposed a novel frame to perform risk stratification and identify PCa risk-related genes via integration of PCa omics data from The Cancer Genome Atlas (TCGA). Firstly, risk genes were extracted by applying Cox regression model. Secondly, consensus non-negative matrix factorization (CNMF) cluster algorithm was applied to RNA-seq expression data of these risk genes, and PCa patients were divided into two subtypes (named as low risk and high risk respectively). Thirdly, by combining with survival analysis and differential expression analysis based on two subtypes, a PCa subtype-related network module was identified.
Results: The identified network module can serve as biomarkers such as SRC to predict PCa risk. In particular, we observed the obvious differences in the DNA methylation profile and copy number variation (CNV) of genes involved in the module between two PCa subtypes.
Conclusions: The framework proposed in this paper provides an effective strategy for the comprehensive analysis of TCGA omics data and can help highlight the prevention and treatment stratification for PCa patients.
