Novel Statistical Network Methodology to Identify and Analyze Cancer Biomarkers

• Activity Number: 185
• Type: Contributed
• Date/Time: Monday, August 4, 2014 : 10:30 AM to 11:15 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Thomas Bartlett*+ and Sofia Olhede and Alexey Zaikin
• Companies: and University College London and University College London
• Keywords: Statistical network models ; Network community detection ; Multivariate statistics ; Cancer biomarkers ; Survival analysis ; Epigenetics

Abstract

The global burden of cancer is expected to increase between 2008 and 2030 from 13 to 22 million new cases each year. The strain that this will put on health services around the world will be reduced by the development of better methods for early detection of disease risk and progression. The epigenome is thought of as the interface between the genome and the environment; hence measurements of DNA methylation, an epigenetic pattern, can indicate exposure to environmental risk factors. However these measurements are extremely noisy, making it a challenge to derive meaningful statistics from such data. Using canonical correlation analysis we have developed a novel statistical measure of the level of interaction between a pair of genes (network nodes) in a single sample/patient, based on DNA methylation data. Testing this interaction measure for association with patient outcome, we show how to construct prognostic networks for cancer, in which the presence of a network edge indi- cates that the network interaction between the corresponding pair of genes (nodes) is statistically significantly prognostic. Detecting community structure in these networks by fitting the stochastic blockmodel allows novel cancer biomarkers to be detected. These findings represent new statistical tools for use in the biomedical sciences.