Identification of disease-specific bio-markers through network-based analysis of gene co-expression: A case study on Alzheimer's disease

Abstract

Finding biomarker genes for complex diseases attracts persistent attention due to its application in clinics. In this paper, we propose a network-based method to obtain a set of biomarker genes. The key idea is to construct a gene co-expression network among sensitive genes and cluster the genes into different modules. For each module, we can identify its representative, i.e., the gene with the largest connectivity and the smallest average shortest path length to other genes within the module. We believe these representative genes could serve as a new set of potential biomarkers for diseases. As a typical example, we investigated Alzheimer's disease, obtaining a total of 16 potential representative genes, three of which belong to the non-transcriptome. A total of 11 out of these genes are found in literature from different perspectives and methods. The incipient groups were classified into two different subtypes using machine learning algorithms. We subjected the two subtypes to Gene Ontology analysis and Kyoto Encyclopedia of Genes and Genomes analysis with healthy groups and moderate groups, respectively. The two sub-type groups were involved in two different biological processes, demonstrating the validity of this approach. This method is disease-specific and independent; hence, it can be extended to classify other kinds of complex diseases.

Keywords

• Gene co-expression network
• Machine learning
• Alzheimer's disease subtypes