Analyzing breast cancer comorbidities: a network approach using community detection algorithms

Abstract

Abstract Breast cancer is a prominent global health concern, as the data from the International Agency for Research on Cancer (IARC) shows that breast cancer is the leading cancer type with new cases in 2020 and among the Top 5 cancer types with the most deaths. To help improve the current breast cancer comorbidity identification by medical personnel and ultimately, lower the number of death cases from breast cancer comorbidity, this research aims to discover the breast cancer comorbidity community, do modularity and similarity-based evaluation, suggest the best semantic similarity measurement and threshold value, and validate the data of breast cancer comorbidities with several data from research papers. The Wang algorithm, with a threshold value of 0.5, is chosen to build the network. Leiden, Louvain, RBER Pots, RB Pots, and Walktrap are the best five community detection algorithms. Similarity measurements with the best three fitness functions (edges inside, scaled density, and size) suggest that the Leiden–Louvain algorithm and RBER Pots-RB Pots algorithm are two pairs of algorithms with similar results. Other similarity measurements with the V-measure heatmap suggest that Louvain–Leiden (0.99), RB Pots–Leiden (0.97), and RB Pots–RBER Pots (0.96) results are similar. Comorbidity is then evaluated using the best five community detection algorithms and four centrality algorithms. As a result, fourteen diseases are agreed upon by the best five community detection algorithms, five diseases are agreed by four algorithms, two diseases are agreed by three algorithms, a disease is agreed by two algorithms, and ten diseases are agreed by an algorithm.

Keywords

• Bioinformatics
• Centrality
• Clustering
• Disease
• Fitness function
• Graph analytics