Table 1.

GAW20 quality control and statistical models, data sets, and software used by this group

Contribution	Phenotype	Normalization	h2	Covariates	CpG probes	Model(s)	Software
Almeida et al. [22]	HDL	Inverse	Pre and post fenofibrate HDL and CpG sites	20 PCs	Epigenome-wide	VC-LME	SOLAR
Canty and Paterson [20]	TG, QC	Probe type strata	–	4 PCs Difference in TGs	Epigenome-wide	Standard linear model (t-test)
Fernandez-Rhodes et al. [25]	Metabolic syndrome	Type II probes	Metabolic syndrome, 4 CpG sites	age, sex, SNPs, center, smoking, PCs	4 CpG sites	VC-LME	SOLAR
LeBlanc et al. [21]	QC only	BMIQ	Used breeding values from a heritability model	Age, sex, SNPs	Epigenome-wide	Bayesian LME	R-INLA
Lim et al. [23]	TG	BMIQ		Age, sex, study, center, smoking, 10 PCs	14,850 CpG sites showing	LME	WGCNA, missmethyl
Nustad et al. [24]	TG, HDL	BMIQ	P-to-e and post fenofibrate, TG HDL and CpG sites	Age, sex	Epigenome-wide	Bayesian LME	R-package INLA
Yu et al. [26]	TG	–		Age, sex, study center, smoking status, HDL	349,755 CpG sites	GEE	R 3.2

All contributions in this group used the GAW20 real data set

BMIQ beta-mixture quantile normalization method for correcting probe design bias, GEE generalized estimating equation, HDL high-density lipoprotein, h² denotes heritability and indicates if the paper has estimated this quantity, INLA integrated nested Laplace approximation, PC DNA methylation-derived principal component and indicates this study employed PCs as covariates in their analysis, QC quality control, SOLAR sequential oligogenic linkage analysis routines, SNP single-nucleotide polymorphism, TG triglyceride, VC-LME variance component linear mixed effect, WGCNA weighted gene coexpression network analysis