Table 2. Coefficient of determination (R 2) estimated using 10-fold cross validation of our adaptive prediction algorithm.
Method | mFDR cutoff | BP | SNP effects | cutoff fraction of occurrence | |||||||
0.01 | 0.05 | 0.1 | 0.2 | ||||||||
Non-SNP first | 0.2 | adjusted | non-SNP | 0.270 | (0.014) | 0.271 | (0.014) | 0.261 | (0.015) | 0.244 | (0.015) |
main | −0.007 | (0.003) | −0.009 | (0.004) | −0.003 | (0.003) | −0.005 | (0.002) | |||
inter | 0.001 | (0.002) | −0.003 | (0.004) | 0.004 | (0.003) | 0.001 | (0.002) | |||
unadjusted | non-SNP | 0.146 | (0.011) | 0.141 | (0.011) | 0.143 | (0.011) | 0.130 | (0.012) | ||
main | −0.002 | (0.002) | 0.004 | (0.003) | 0.002 | (0.002) | 0.000 | (0.003) | |||
inter | −0.001 | (0.003) | 0.004 | (0.003) | −0.007 | (0.003) | −0.012 | (0.005) | |||
0.5 | adjusted | non-SNP | 0.273 | (0.014) | 0.270 | (0.014) | 0.260 | (0.015) | 0.244 | (0.014) | |
main | 0.220 | (0.020) | 0.230 | (0.019) | 0.232 | (0.015) | 0.194 | (0.012) | |||
inter | 0.248 | (0.013) | 0.250 | (0.013) | 0.238 | (0.014) | 0.227 | (0.012) | |||
unadjusted | non-SNP | 0.170 | (0.011) | 0.170 | (0.011) | 0.171 | (0.011) | 0.151 | (0.013) | ||
main | 0.128 | (0.010) | 0.133 | (0.011) | 0.128 | (0.013) | 0.110 | (0.012) | |||
inter | 0.170 | (0.011) | 0.171 | (0.011) | 0.172 | (0.011) | 0.147 | (0.014) | |||
SNP first | 0.2 | adjusted | SNP only | 0.133 | (0.010) | 0.132 | (0.010) | 0.133 | (0.010) | 0.134 | (0.010) |
non-SNP main | 0.268 | (0.013) | 0.264 | (0.012) | 0.264 | (0.012) | 0.260 | (0.013) | |||
non-SNP inter | 0.271 | (0.014) | 0.268 | (0.014) | 0.261 | (0.016) | 0.249 | (0.014) | |||
Non-SNP first+candidate SNPs | 0.2 | adjusted | non-SNP | 0.270 | (0.014) | 0.271 | (0.014) | 0.261 | (0.015) | 0.244 | (0.015) |
SNP main | 0.263 | (0.014) | 0.264 | (0.014) | 0.259 | (0.016) | 0.241 | (0.016) | |||
SNP inter | 0.269 | (0.014) | 0.269 | (0.014) | 0.261 | (0.015) | 0.248 | (0.015) |
“Non-SNP first”: the non-SNP features were first selected and the main effects of SNPs were chosen at the marginal false discovery rate cutoff of 0.2 and 0.5.
“SNP first”: the main effects of SNPs were first selected at the marginal false discovery rate cutoff of 0.2 and non-SNP effects were later included.
“candidate SNPs”: the non-SNP features were first selected and the 26 candidate SNPs were included together with the main effects of SNPs that were chosen at the marginal false discovery rate cutoff of 0.2.
For the column “BP”, “adjusted” is for results for medication-adjusted mean arterial blood pressure, and “unadjusted” for unadjusted blood pressure for each cutoff fraction of occurrence in the bootstrapped CART.
For each method and mFDR cutoff, the first row presents the baseline R 2; “main” and “inter” refers to the increase or decrease in R 2 from “non-SNP.” Standard errors of the individual R 2 for each of the ten folds are presented within parentheses.