Table 1 Paper, data used, and methods implemented for capturing genetic, epigenetic, and associations with phenotypes
From: The challenge of detecting genotype-by-methylation interaction: GAW20
No | Paper | Data used | GAW20 simulations Answers known | Methods | Specific data and modeling | Response variable |
|---|---|---|---|---|---|---|
1 | Daw et al | Simulated | YES | GLM Linear mixed models Regression trees Random forest | Simulated data (200 replications) Simulated effect sites + noise | ave(log(TGpost)) ave.(log(TGpost)) − ave.(log(TGpre)) |
2 | Fisher et al | Real data | N/A | Mediation interaction model | GWAS EWAS pairs with P1 < 10−3 and P2 < 0.05 The effect of change in methylation on change in TG differs by genotype Power: simulation with 500 replications | logTGpost − logTGpre |
3 | Hu and Li | Real data | N/A | Gaussian mixture-model clustering Gaphunter a thresholding-based method | CpG × SNP (GW) multimodal CpG sites Association of CpG identified with overlapping SNP Association with nearby SNPs | Pretreatment CpG Posttreatment CpG |
4 | Romanescu et al | Real data | N/A | Regional-based test | Single CpG association (CH11) Regional SNP PC association (CH11) Regional CpG PC association (CH11) Regional SNP + regional CpG association (CH11) | ave(ln(TGpre)) |
5 | Sun et al | Both | YES | Adaptive W-test with 2000 repetitions to get Type I error | In both analysis main and interaction associations Simulation: candidate pairs and noise pairs Real data: (CH11) | delta(TG%) = 1, if delta(TG%) > 30% = 1, otherwise 0 delta(TG%) = (TGpre − TGpost)/TGpre |
6 | Veenstra et al | Real data | N/A | Lmekin of coxme, lm in R | (1) SNP association + (2) CpG association (1) + (2) + CpG × SNP | Ave(TG1 + TG2) baseline HDL |
7 | Zhou and Lo | Real data | N/A | Group LASSO-regression, kinship matrix, stability selection | CpG association (GW) SNP association (GW) (1) + (2) + CpG × SNP (CH11) | Ave(TG) |