Table 2 The experiment result on the AMD data set
Model | SNP name | CV accuracy | CV consistency | χ2-value |
|---|---|---|---|---|
M 1 | rs1329428 | 0.7246 | 10/10 | 27.1480 |
M 2 | rs1329428, rs9299597 | 0.7086 | 6/10 | 38.3007 |
M 3 | rs1535891, rs1329428, rs9299597 | 0.7833 | 3/10 | 55.4297 |
M 4 | rs1535891, rs2828151, rs404569, rs380390 | 0.7615 | 10/10 | 85.2449 |
- The result obtained by using MDR after screening. ReliefF is used to reduce the number of SNPs from L = 82, 143 to d = 20. After that, MDR is applied to the remaining 20 SNPs to search for the best model. The fourth-way interaction model M4 is considered as the best model because it gives high CV accuracy and CV consistency. Here CV stands for cross-validation. The χ2-value of 85.2449 will correspond to a P-value of 4.226 × 10–7 against the null distribution
(the upper panel of Figure 2 (d = 4)), indicating a significant hypothesis testing result. But the permutation test gives it a P-value of 0.1480 (Figure 5), indicating a non-significant result.
(the upper panel of Figure