Skip to main content

Table 1 Summary statistics and variable importance rankings of simulated causal effects on TG

From: Using recursive feature elimination in random forest to account for correlated variables in high dimensional data

  

RF*

RF-RFE

r2

−0.00203

0.19217a

MSEOOB

0.07378

0.05948a

Chr

Causal SNP/CpG

MAF or Mean (SD)

Simulated h2†

Main Effects, β(SE), p value

Interaction Effects, β(SE), p value

Rank (percentile rank)

1

rs9661059

0.12

0.125

.14 (0.02), <0.0001

−0.19 (0.07), 0.0109

1 (100.0)

20 (100.0)

cg00000363

0.49 (0.33)

−0.05 (0.03), 0.0873

8680 (97.6)

239,755 (32.7)

6

rs736004

0.09

0.075

0.09 (0.03), 0.0005

−0.30 (0.08), 0.0001

13,480 (96.2)

766 (99.8)

cg10480950

0.54 (0.33)

−0.04 (0.03), 0.1711

5332 (98.5)

232,579 (34.7)

8

rs1012116

0.20

0.100

0.08 (0.02), <0.0001

−0.21 (0.06), 0.0002

50,218 (85.9)

333,504 (6.4)

cg18772399

0.56 (0.33)

−0.002 (0.03), 0.9466

339,475 (4.7)

301,855 (15.3)

10

rs10828412

0.14

0.025

0.07 (0.02), 0.0007

−0.07 (0.07), 0.2770

2984 (99.2)

330,516 (7.3)

cg00045910

0.49 (0.34)

0.01 (0.03), 0.7176

263,465 (26.1)

231,315 (35.1)

17

rs4399565

0.41

0.050

0.04 (0.02), 0.0142

−0.13 (0.05), 0.0038

11,078 (96.9)

196,276 (44.9)

cg01242676

0.46 (0.32)

0.01 (0.03), 0.8159

350,420 (1.9)

350,420 (1.7)

  1. Bolded values are significant at a p<0.05
  2. *RF is the first RF in RF-RFE
  3. ar2 and MSEOOB are averaged over all 324 RFs in the RF-RFE column
  4. Simulated h2 was provided by GAW20 organizers and based on full 200 simulations; main and interaction effects are calculated within the data set used for this study, which uses the 84th simulation replicate. Effects are calculated with linear regression models using the residual of change in TG after adjusting for baseline TG as the outcome. Interaction effects include the main effects of the interaction terms being tested in the given model
  5. Abbreviations: β Effect size, Chr Chromosome, h2 Heritability