Guidance for DNA methylation studies: statistical insights from the Illumina EPIC array.

Mansell G, Gorrie-Stone TJ, Bao Y, Kumari M, Schalkwyk LS, Mill J, Hannon E. · 2019

Read Guidance for DNA methylation studies: statistical insights from the Illumina EPIC array. by Mansell G, Gorrie-Stone TJ, Bao Y, Kumari M, Schalkwyk LS, Mill J, Hannon E. (2019) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc6518823.txt.

Introduction

Mansell and co-authors give guidance for DNA methylation studies on the Illumina EPIC array. Their paper appeared in BMC Genomics. It tackles three questions of study design and analysis. How strict should significance thresholds be? Is linear regression sound for this data? And how large must samples be?

The team measured methylation in the Understanding Society cohort. That cohort held 1,175 people. Null simulations set a 5% family-wise error threshold of 9 × 10⁻⁸. Most sites broke the normal-residual assumption of linear regression. Yet the team found no sign that this raised false positives. Power checks showed about 1,000 samples can detect small differences at most sites.

Catalog record

CREATORMansell G, Gorrie-Stone TJ, Bao Y, Kumari M, Schalkwyk LS, Mill J, Hannon E.
YEAR2019
COLLECTIONPAPERS
FORMATTXT
FILE SIZE67 KB
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC6518823

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Guidance for DNA methylation studies: statistical insights from the Illumina EPIC array By Georgina Mansell, Tyler J. Gorrie-Stone, Yanchun Bao, Meena Kumari, Leonard S. Schalkwyk, Jonathan Mill, Eilis Hannon Published in: BMC Genomics DOI: 10.1186/s12864-019-5761-7 ABSTRACT Background There has been a steady increase in the number of studies aiming to identify DNA methylation differences associated with complex phenotypes. Many of the challenges of epigenetic epidemiology regarding study design and interpretation have been discussed in detail, however there are analytical concerns that are outstanding and require further exploration. In this study we seek to address three analytical issues. First, we quantify the multiple testing burden and propose a standard statistical significance threshold for identifying DNA methylation sites that are associated with an outcome. Second, we establish whether linear regression, the chosen statistical tool for the majority of studies, is appropriate and whether it is biased by the underlying distribution of DNA methylation data. Finally, we assess the sample size required for adequately powered DNA methylation association studies. Results We quantified DNA methylation in the Understanding Society cohort (n = 1175), a large population based study, using the Illumina EPIC array to assess the statistical properties of DNA methylation association analyses. By simulating null DNA methylation studies, we generated the distribution of p-values expected by chance and calculated the 5% family-wise error for EPIC array studies to be 9 × 10− 8. Next, we tested whether the assumptions of linear regression are violated by DNA methylation data and found that the majority of sites do not satisfy the assumption of normal residuals. Nevertheless, we found no evidence that this bias influences analyses by increasing the likelihood of affected sites to be false positives. Finally, we performed power calculations for EPIC based DNA methylation studies, demonstrating that existing studies with data on ~ 1000 samples are adequately powered to detect small differences at the majority of sites. Conclusion We propose that a significance threshold of P < 9 × 10− 8 adequately controls the false positive rate for EPIC array DNA methylation studies. Moreover, our results indicate that linear regression is a valid statistical methodology for DNA methylation studies, despite the fact that the data do not always satisfy the assumptions of this test. These findings have implications for epidemiological-based studies of DNA methylation and provide a framework for the interpretation of findings from current and future studies. Electronic supplementary material The online version of this article (10.1186/s12864-019-5761-7) contains supplementary material, which is available to authorized users. BACKGROUND There is increasing interest in the role of epigenetic processes in health and disease, with the primary focus of most epigenetic epidemiological studies being on DNA methylation (DNAm) [1]. Platforms such as the Illumina 450 K Human Methylation microarray (450 K array) and the Illumina EPIC Human Methylation microarray (EPIC array) have enabled the economical, high-throughput profiling of methylomic variation across large numbers of samples. In recent years a number of epigenome-wide association studies (EWAS), which aim to identify DNAm differences associated with environmental exposure and disease, have been reported for a range of complex phenotypes including cancer [2–4], autoimmune disorders [5–7], psychiatric illnesses [8, 9], neurodevelopmental disorders [10, 11] and dementia [12, 13]. Primarily due to the dynamic nature of the epigenome throughout development, across different cell types and in response to environmental exposures, much has previously been written regarding the specific nuances of performing an EWAS compared to a genome-wide association study (GWAS) of genetic variation [14–16]. However, this literature is mainly focused on study design and interpretation rather than specific analytical issues relating to the characteristics of the data. One concern that has merited some discussion relates to whether the distribution of DNAm data violates the assumptions of Gaussian linear regression [17, 18], the most commonly used analysis model as it allows for the inclusion of covariates relating to both biological and technical confounders. For each molecule of DNA in a single cell, DNAm is a binary entity, in that at any cytosine it is either present or absent. However, as almost all DNAm studies profile either bulk tissue - comprising multiple cell types - or a population of purified cells, the analyses are essentially measuring the proportion of cells (taking a value between 0 and 1) in a sample that are methylated at a specific genomic position [19]. While across the sites profiled on Illumina arrays DNAm has a bimodal distribution with peaks of hypomethylation (i.e. unmethylated sites) and hypermethylation (i.e. methylated sites), there is a significant subset of sites exhibiting intermediate levels of DNAm (proportion of methylated alleles = ~ 0.5). As the presence/absence of DNAm primarily distinguishes different cell types and tissues, in studies of a single tissue, which the majority of epigenetic epidemiology studies are, it is unlikely that the distribution at individual DNAm sites (the standard unit of analysis in an EWAS) will be bimodal. However, it is likely that the distributions will be variable and often non-normal, meaning that the assumption that the residuals of a linear regression fit are normally distributed may not hold. Furthermore, as DNAm levels are bounded by the limits of 0 and 1 it means that at the extreme ends of the distribution the variance is compressed. States of hypo and hypermethylation often define cell types and would not be expected vary biologically within a cell type, beyond any technical noise in the assay. This is exacerbated by the fact that the sensitivity of the microarray technology is less precise at these extremes of the distribution, and hence some measured variation is often present for these theoretically non-variable sites. This property of the data is called heteroskedasticity, defined as a relationship between the mean and variance of a dataset, and violates another assumption of linear regression. Although these concerns should be considered when it comes to deciding the statistical methodology, it is not currently known whether these violations are sufficient to bias analyses and introduce false positive or even false negative findings. Consistent with studies of other types of genomic variation, another challenge for EWAS is how to account for the multiple testing burden in a typical analysis; for example, the Illumina EPIC array assays DNAm at base pair resolution for > 850,000 sites across the genome. Currently, a range of approaches are used to establish an appropriate significance threshold and there is no standard significance threshold as is used in GWAS. A common approach is a Bonferroni correction for the number of probes on the array [20–23] although this is often presumed to be too conservative as DNAm values at neighboring probes are known to be correlated [24], and many sites on the array are non-variable. An alternative, potentially more powerful, approach sets a permissible false discovery rate (FDR), and identifies the top associated sites that satisfy this criterion [25]. While FDR can be calculated by generating the empirical null distribution of test statistics [26], it is most commonly applied using the approach introduced by Benjamini and Hochberg [27]. This makes the assumption that under the null hypothesis the p-values across individual sites are uniformly distributed [28], which is not necessarily true. In EWAS it is not uncommon to see inflated test statistics [29, 30], even in the scenario of

Read the complete text