Global Spectral Clustering in Dynamic Gene Co-Expression Networks

• Activity Number: 513 - Gene Expression Analysis
• Type: Contributed
• Date/Time: Wednesday, August 2, 2017 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Genomics and Genetics
• Author(s): Fuchen Liu* and Kathryn Roeder and David Choi and Lu Xie
• Companies: Carnegie Mellon University and Carnegie Mellon University and Carnegie Mellon University and Carnegie Mellon University
• Keywords: spectral clustering ; community detection ; dynamic networks ; gene co-expression ; cross validation ; eigengap

Abstract

Recently, the gene co-expression networks are increasingly being used for finding gene correlated communities and linking candidate risk genes. The changes of gene expressions across brain developing periods are also well analysised. However, the changing of gene communities in developing brains remains poorly understood. The biggest obstacle is the lack of samples: the sample size would become even smaller when we divided them into different time periods. We proposed a global spectral clustering method which combines common information of a series of networks to strengthen the inference of each time period. A data-driven extended eigengap method is also provided to estimate the number of clusters. To measure the method performance and apply model selection, we also proposed a cross validation procedure for adjacency matrix. We implement our method in a recently published rhesus monkey data. Gene communities paths and annotation terms are identified to illustrate the community changing across time.