Adaptive Resistant Regression Method (ARM): A Better Alternative to Quantile Normalization for Methylation Data

• Activity Number: 303
• Type: Contributed
• Date/Time: Tuesday, August 6, 2013 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Jean-Philippe Fortin*+ and Aurélie Labbe and Celia M.T. Greenwood and Mathieu Lemire and Brent W. Zanke and Thomas J. Hudson
• Companies: Department of Biostatistics, Johns Hopkins School of Public Health and Department of Epidemiology, Biostatistics and Occupational Health, McGill University and Department of Epidemiology, Biostatistics and Occupational Health, McGill University and Ontario Institute of Cancer Research, Toronto, ON, Canada and Ottawa Hospital Research Institute and Ontario Institute of Cancer Research, Toronto, ON, Canada
• Keywords: Normalization ; Microarray ; DNA Methylation ; Epigenetics ; Illumina 450k ; Genomics

Abstract

Motivation: DNA methylation levels can be estimated with microarrays, but the signals require normalization to adjust for technical artifacts. Current packages for Illumina Infinium HumanMethylation 450k platform use different versions of quantile normalization, but none is completely appropriate for methylation data.

Results: We present a new method, the Adaptive Resistant Regression method (ARRm) for normalization of DNA methylation data from the Illumina Infinium HumanMethylation450 BeadChip, using a one-step correction model for background noise, dye bias, design bias and spatial effects. It is a novel approach containing adjustments that vary with the percentiles of the methylation levels and taking advantage of the negative control probes. The key feature of the method is that contrary to methods based on quantile normalization, it does not rely on an assumption of distributional similarity across subjects. ARRm is carefully evaluated against several other approaches for normalization, using a dataset containing 2053 samples where methylation was measured in either lymphocytes or lymphoblasts. Our method is shown to be the most appropriate choice for DNA methylation data.