Controlling bias and inflation in epigenome- and transcriptome-wide association studies using the empirical null distribution

Abstract

We show that epigenome- and transcriptome-wide association studies (EWAS and TWAS) are prone to significant inflation and bias of test statistics, an unrecognized phenomenon introducing spurious findings if left unaddressed. Neither GWAS-based methodology nor state-of-the-art confounder adjustment methods completely remove bias and inflation. We propose a Bayesian method to control bias and inflation in EWAS and TWAS based on estimation of the empirical null distribution. Using simulations and real data, we demonstrate that our method maximizes power while properly controlling the false positive rate. We illustrate the utility of our method in large-scale EWAS and TWAS meta-analyses of age and smoking.Electronic supplementary materialThe online version of this article (doi:10.1186/s13059-016-1131-9) contains supplementary material, which is available to authorized users.

Schizophrenia-associated methylomic variation: molecular signatures of disease and polygenic risk burden across multiple brain regions

“…S10A–D highlighting some evidence of P- value inflation (PFC λ = 1.18, STR λ = 1.02, HC λ = 1.13, CER λ = 1.23) in several of the brain regions; such inflation is not unusual in epigenome-wide association study (EWAS) analyses and standard genomic control methods – widely used in GWAS – are not suitable for EWAS data (22). Because it is likely that unmeasured factors beyond the variables included in our analysis model (i.e.…”

“…Although smoking has been shown to have striking effects on DNA methylation in blood (52), none of the robust smoking-associated DMPs identified in the blood are amongst the schizophrenia DMPs identified in any of the four brain regions assessed in the current study. Although P -value inflation is a common feature of many DNA methylation datasets, standard genomic control methods – widely used in GWAS – are not suitable for EWAS data (22). Therefore, we investigated the impact of including additional surrogate variables capturing variation in DNA methylation on the association statistics for schizophrenia-associated DMPs (Supplementary Material, Figs S11–S14), observing that the identified schizophrenia-associated DMPs are relatively robust to the major PCs associated with methylomic variance.…”

Integrated Analysis of Genetic Ancestry and Genomic Alterations across Cancers

“…Q-Q plots were generated by the function stat_qq() of R’s package ggplot2. Moreover, we calculated inflation factors using the R/Bioconductor package BACON (van Iterson et al, 2016) to assess whether inflation was introduced by the analysis.…”

DRD4 methylation as a potential biomarker for physical aggression: An epigenome‐wide, cross‐tissue investigation

“…The results of the probe‐level EWAS are depicted in Table . Overall, inspection of the Q‐Q plot indicated little evidence of inflation (Supporting Information Figure S1), which was confirmed using the Bayesian method in the Bacon R package (van Iterson, van Zwet, & Heijmans, ; Supporting Information Figure S2). Four DMPs were found to meet a genome‐wide threshold of significance ( q ≤ .05; bolded in Table ; see Supporting Information Figure S1 for the Manhattan plot and scatterplots for each DMP): (a) cg21217577, located in the body of the TRH gene, a key member of the hypothalamic–pituitary–thyroid axis involved in hormonal regulation (Yarbrough, Kamath, Winokur, & Prange, ); (b) cg11107262 in the body of GAK , a gene involved in cellular function (Shimizu, Nagamori, Yabuta, & Nojima, ); (c) cg27349081, annotated to the promoter region of RIC3 , a molecular chaperone modulating the expression of nicotinic acetylcholine and serotoninergic 5‐HT3 receptors (Castillo et al, ); and (d) cg27331554 located in the body of CYFIP2 , a component of the WAVE1 complex implicated in immune function and BDNF‐NTRK2 signaling (Xu, Fu, Zhu, & Liu, ).…”