EMODMI: A Multi-Objective Optimization Based Method to Identify Disease Modules

After decades of research, it has been widely recognized that complex diseases are caused by the dysfunction of biological systems induced by disease-associated genes. To understand the molecular basis of complex diseases, many efforts have been devoted to the identification of disease-related gene modules in the last two decades, by means of exploring the interaction networks constructed based on heterogeneous information. However, many existing approaches ignore the personalized features of disease samples and cannot identify a dense module having strong association with the disease. In this paper, an evolutionary multi-objective optimization based approach is proposed for disease module identification. The proposed approach constructs a sample-specific network for each disease sample to involve their personalized features, then optimizes both the association of the module with the disease and the intra-link density of the module by using a multi-objective genetic algorithm. According to the experimental results on the asthma gene expression dataset, the proposed approach is superior over some state-of-the-art disease module identification approaches. Furthermore, the identified disease module is used in the classification of disease and control samples, which obtains lower classification error rate than existing approaches.

Automated summary

This paper proposes an evolutionary multi-objective optimization approach for disease module identification, which constructs sample-specific networks to involve personalized features, and optimizes module association with disease and intra-link density using a genetic algorithm. The approach outperforms existing methods in identifying disease modules and leads to lower classification error rates in disease classification experiments.