A-clustering: A Novel Method for the Detection of Methylation Regions Associated with Exposure

摘要

Background: High-throughput methylation data have become increasingly abundant. There is a growing interest in powerful and efficient methods for detection of methylation regions associated with environmental exposure. Aims: Develop a method to identify regions of co-regulated methylation sites and detect their association with exposure. Methods: We develop the A-clustering algorithm that identifies sets of neighboring CpG sites that are co-regulated. We provide a software package that efficiently implements this algorithm, and performs a robust GEE-based analysis to study the association of the detected methylation regions with exposure. We examine the A-clustering and the testing approach in a correlation-preserving simulation study, and apply the method on 450K Infinium BeadChip data from peripheral blood of eighty pesticide applicators from a NIH-funded project studying genomewide DNA methylation patterns in response to pesticide exposure. Results: A-clustering and the proposed analysis pipeline are quick, robust, and perform well in simulations. Many regions of co-regulated methylation sites in the pesticides applicators data set did not correspond to known functional domains such as CpG islands. Similarly, 34% of the regions associated with exposure were not related to known functional domains. Conclusions: Regions of co-regulated CpG sites are broader than the pre-defined domains such as CpG islands and shores. The A-clustering is useful to detect such regions and their relation to environmental exposures. The proposed analysis pipeline is powerful and more efficient than other existing methods.