Table 4.
Mean prediction performance of genes predicted in METS by all 4 of the prediction algorithms for each training population
| Population | EN | RF | SVR | KNN | Number of genes |
|---|---|---|---|---|---|
| AFA | 0.1210 | 0.1150 (p = 3.0 × 10−3) | 0.0959 (p = 4.5 × 10−23) | 0.0723 (p = 1.9 × 10−47) | 1,640 |
| HIS | 0.0880 | 0.1066 (p = 5.3 × 10−11) | 0.0896 (p = 5.9 × 10−1) | 0.0648 (p = 1.4 × 10−12) | 1,809 |
| CAU | 0.0620 | 0.0770 (p = 1.3 × 10−7) | 0.0699 (p = 4.0 × 10−3) | 0.0475 (p = 5.4 × 10−6) | 2,091 |
| AFHI | 0.1111 | 0.1068 (p = 1.1 × 10−2) | 0.0944 (p = 8.8 × 10−16) | 0.0695 (p = 9.6 × 10−50) | 2,290 |
| ALL | 0.1074 | 0.1046 (p = 1.2 × 10−1) | 0.0944 (p = 1.3 × 10−11) | 0.0659 (p = 9.0 × 10−49) | 2,315 |
For each training population, we took only intersection genes predicted by EN, RF, SVR, and KNN. Focusing on these intersects for each training population, we calculated the mean prediction performance (ρ). The paired t test p value between EN and each other model is shown in parentheses. EN, elastic net; RF, random forest; SVR, support vector regression; KNN, K nearest neighbor; AFA, MESA African American; HIS, MESA Hispanic American; CAU, MESA European American; AFHI, MESA African American and Hispanic American; ALL, all MESA.