A-clustering: a novel method for the detection of co-regulated methylation regions, and regions associated with exposure

Abstract

tsofer@hsph.harvard.edu

Lead exposure induces changes in 5-hydroxymethylcytosine clusters in CpG islands in human embryonic stem cells and umbilical cord blood

“…20 Houseman et al, 2012 proposed a statistical model for estimation of blood cell type proportion using methylation data. 61 Using this methodology, they found that the leukocyte distribution accounts for only 3% of the differences in the DNA methylation profile of individuals exposed to arsenic (As).…”

“…The co-regulated regions can be assigned to specified clusters and the effect of the exposure on these clusters can be tested using the generalized estimating equation (GEE). 20 GEE uses a weighted combination of observation to measure the effect of a covariate (in our case Pb exposure) while conserving the correlation structure of the data. 64 Consequently this approach is much less conservative and yields a greater number of differentially methylated regions compared to the traditional case-control study with a single CpG site b value comparison.…”

“…19 These regions often exhibit similar behavior in response to environmental stimuli and can be assigned to specific clusters or co-regulated regions on the basis of a correlation between their methylation signatures and bases pair distance. 20 Therefore, in this study, we first define co-regulated regions of the genome, independent of any exposure or classifier data and then fit a generalized estimating equation to measure the effect of exposure on these regions. This approach, called adjacent site clustering (A-clustering), was proposed by Sofer et al, 2012 and has been shown to perform significantly better than a similar method called "bump hunting".…”

“…This approach, called adjacent site clustering (A-clustering), was proposed by Sofer et al, 2012 and has been shown to perform significantly better than a similar method called "bump hunting". 20,21 Furthermore, we believe that studying exposure-associated changes in 5mC and 5hmC as co-regulated regions is more biologically relevant because DNMT1, DNMT3a and TET proteins act on multiple sites in a small region when they bind to DNA. 22 These Pb-targeted co-regulated regions can influence a variety of biological processes including binding of transcription factors and may serve as putative epigenetic biomarkers for Pb exposure.…”

Epigenome-Wide Cross-Tissue Predictive Modeling and Comparison of Cord Blood and Placental Methylation in a Birth Cohort

“…The samples of the training set were used to fit the linear regression model and to compute the predictor coefficients' estimates are, respectively, the cord blood methylation value and the mean of the correlated nearby cord blood methylation sites from the sample being predicted. The assign.to.clusters function in the R package Aclust [19] was used to define the sets of neighboring CpG sites that are correlated with each other, using default parameters. This function implements a clustering method to discover methylation regions by clustering together adjacent probes within a genomic distance constraint according to their Spearman correlation between samples (within a single tissue).…”