Table 1.
Accuracy of iMAP in selecting informative annotations and in estimating the annotation coefficients in the setting where four independent annotations with relatively large effects are present
Proportion of pleiotropic causal SNPs (%) | True | False | MSE |
||
---|---|---|---|---|---|
Oracle | Select | Full | |||
0 | 3.53 | 0.38 | 0.05 | 0.28 | 383.52 |
20 | 7.81 | 0.53 | 0.41 | 2.38 | 300.15 |
40 | 9.72 | 1.20 | 0.29 | 2.03 | 68.63 |
60 | 5.77 | 0.81 | 0.31 | 2.58 | 12.53 |
80 | 4.98 | 0.75 | 0.56 | 2.70 | 17.46 |
100 | 3.15 | 0.37 | 0.12 | 2.17 | 18.38 |
Note: Simulations were carried out in the presence of four informative annotations and 100 non-informative annotations for various proportion of pleiotropic causal SNPs (rows). The ‘True’ column lists the number of selected correct non-zero annotation parameters inside the mlogit model. Note that a total of 12 (= 3 × 4) non-zero annotation parameters are expected in the presence of four informative annotations. The ‘False’ column lists the number of selected incorrect non-zero annotation parameters. MSE denotes the median squared error for the estimated annotation parameters across 100 simulation replicates for three different versions of iMAP: the oracle version uses the four informative annotations; the select version performs annotation selection; and the full version includes all annotations without selection.