Improved prediction of breast cancer outcome by identifying heterogeneous biomarkers

• DOI: 10.1093/bioinformatics/btx487
• Publication Date: 2017-07-28T03:10:50Z
• Authors: Jonghwan Choi; Sanghyun Park; Youngmi Yoon; Jaegyoon Ahn
• Keywords: Identification

Abstract Motivation Identification of genes that can be used to predict prognosis in patients with cancer is important it lead improved therapy, and also promote our understanding tumor progression on the molecular level. One common but fundamental problems render identification prognostic prediction outcomes difficult heterogeneity patient samples. Results To reduce effect sample heterogeneity, we clustered data samples using K-means algorithm applied modified PageRank functional interaction (FI) networks weighted gene expression values each cluster. Hub among resulting prioritized were selected as biomarkers This process outperformed traditional feature selection methods well several network-based when Random Forest. We able find many cluster-specific for dataset. Functional study showed distinct biological processes enriched cluster, which seems reflect different aspect or oncogenesis groups. Taken together, these results provide support hypothesis approach effectively identify heterogeneous genes, are complementary other, improving accuracy. Availability implementation https://github.com/mathcom/CPR Supplementary information available at Bioinformatics online.