Table 9.

Comparison of the best performing probes.

	cg04692870-Probe 1, best R2	cg04692870-Probe 1, best RMSE	cg09322432-Probe 3	cg10840135-Probe 4, best R2	cg10840135-Probe 4, best RMSE	cg15597984-Probe 5	cg20046859-Probe 7	Mean of best performance/ RMSE (SD)
Model and feature set	Elastic net Ind GWAS after B-H genetics only	XGBoost Ind GWAS after B-H genetics only	Elastic net Ind GWAS uncorrected genetics + environmental	Elastic net GTEx	XGBoost 2 MB genetic only	Elastic net Ind GWAS uncorrected genetics + environmental	Elastic net Ind GWAS uncorrected genetics only	–
No. of Features	2 SNPs (after 10% MAF filtering) + CpG probe	2 SNPs (after 10% MAF filtering) + CpG probe	190 SNPs (after 10% MAF filtering) + 5 PCs + 8 non-genetic factors	548 SNPs (after 10% MAF filtering) + CpG probe	2,406 SNPs (after 10% MAF filtering) + CpG probe	170 SNPs (after 10% MAF filtering) + 5 PCs + 8 non-genetic factors	153 SNPs (after 10% MAF filtering) + CpG probe	–
No. trained	300	300	300	302	305	301	304	–
Hyper parameters	alpha: 0 lambda: 0	objective: reg:squared error eta: 0.003009989 max_depth: 10 colsample_ bytree: 0.5493245 subsample: 0.8383967 nround: 1000	alpha: 0 lambda: 0	alpha: 0 lambda: 1	objective: reg:squared error eta: 0.006558051 max_depth: 1 colsample_ bytree: 0.7361971 subsample: 0.7592239 nround: 924	alpha: 0 lambda: 0	alpha: 0 lambda: 0	–
RMSE train set	0.0646	0.0639	0.0185	0.0249	0.0221	0.0279	0.0152	–
R2 train set	0.104	0.099	0.390	0.109	0.286	0.340	0.327	–
No. tested	101	102	101	102	100	100	102	–
RMSE test set	0.0582	0.0580	0.0176	0.0237	0.0237	0.0256	0.0161	0.0282 (0.017)
R2 test set	0.206	0.179	0.107	0.187	0.155	0.302	0.119	0.184 (0.078)