Network-based Modeling for Breast Cancer Classification using Molecular Data

Abstract

Cancer is a disease characterized by uncontrolled cell growth. One of the characteristics of uncontrolled growth is the presence of estrogen-receptor-positive (ER+). About 67% of breast cancer test results have ER+. Breast cancer profiles are divided into 4 subtypes, namely: Luminal A, Luminal B, basal-like, and HER-2 enriched. Each category has a different effect on adjuvant chemotherapy. In this study, a network-based approach was used to select features/molecular biomarkers that have the potential to assist modeling and classifying sub-types of breast cancer. The molecular features used are Copy Number Alteration (CNA) and gene expression. The feature selection results were compared with the PAM50 feature-based accuracy from the literature study. The results indicate that the features selected from this network-based approach can obtain a comparable performance w.r.t the original PAM50 features, and can be used as alternative to perform breast cancer subtyping.