Clique-Based Clustering of Correlated SNPs in a Gene Can Improve Performance of Gene-Based Multi-Bin Linear Combination Test
Table 4
Average MLC test power over all gene-causal-SNP combinations for LDSelect (MLC-LD) and CLQ (MLC-CL) clustering methods and the proportion of genes where MLC-LD power and MLC-CL power are less than Wald test power.
Model
All possible causal SNPs and all genes
All possible causal SNPs for the genes where LDSelect and CLQ clusters are different
% Power < Wald*
% Power < Wald*
LDS
CLQ
LDS
CLQ
LDS
CLQ
LDS
CLQ
A
0.3
11,117
0.627
0.757
36.6
6.2
9,765
0.614
0.759
40.0
5.8
0.4
11,117
0.670
0.758
26.4
3.9
8,867
0.656
0.762
30.5
3.3
0.5
11,117
0.716
0.754
14.6
2.2
8,069
0.714
0.759
17.2
1.8
0.6
11,117
0.735
0.745
6.7
1.0
7,381
0.742
0.753
8.1
0.7
0.7
11,117
0.733
0.730
2.7
0.6
6,234
0.751
0.744
3.4
0.2
0.8
11,117
0.719
0.712
1.1
0.6
5,138
0.746
0.731
0.8
0.0
0.9
11,117
0.691
0.685
1.4
1.3
3,512
0.726
0.707
0.3
0.0
B
0.3
79,650
0.645
0.771
33.7
5.6
74,715
0.640
0.774
35.0
5.2
0.4
79,650
0.682
0.773
25.5
3.6
70,384
0.674
0.775
27.3
3.0
0.5
79,650
0.727
0.769
14.5
2.1
66,788
0.723
0.770
15.8
1.7
0.6
79,650
0.750
0.760
6.4
1.2
63,848
0.752
0.764
7.0
0.9
0.7
79,650
0.748
0.745
3.0
0.6
57,300
0.756
0.752
3.5
0.5
0.8
79,650
0.733
0.724
0.9
0.4
48,577
0.752
0.737
0.7
0.2
0.9
79,650
0.701
0.692
0.9
0.5
33,403
0.724
0.706
0.8
0.1
C
0.3
79,650
0.499
0.649
54.3
23.7
74,710
0.505
0.663
54.2
21.9
0.4
79,650
0.551
0.657
44.1
21.1
70,409
0.557
0.675
44.0
18.6
0.5
79,650
0.603
0.662
32.8
18.4
66,772
0.615
0.683
32.0
15.5
0.6
79,650
0.637
0.664
23.7
16.4
63,910
0.651
0.682
22.8
14.1
0.7
79,650
0.652
0.662
18.1
14.1
57,409
0.669
0.682
17.4
12.2
0.8
79,650
0.654
0.657
14.1
11.7
48,669
0.675
0.678
13.8
10.3
0.9
79,650
0.645
0.646
10.3
8.8
33,625
0.661
0.662
11.6
8.3
D**
0.3
8,883
0.388
0.444
36.5
12.1
7,054
0.372
0.441
39.7
9.9
0.4
8,883
0.408
0.447
28.3
9.7
6,140
0.389
0.440
32.5
7.3
0.5
8,883
0.426
0.447
18.8
7.4
5,119
0.404
0.433
22.1
5.0
0.6
8,883
0.439
0.445
10.8
5.5
4,420
0.425
0.435
12.5
3.5
0.7
8,883
0.439
0.440
6.5
4.2
3,625
0.416
0.414
7.4
3.0
0.8
8,883
0.435
0.433
4.4
3.3
2,827
0.419
0.412
4.1
1.7
0.9
8,883
0.425
0.423
3.6
3.3
2,103
0.406
0.396
2.7
1.7
The differences of power between two clustering algorithm and the proportions of cases with MLC test power less than the power of Wald test within genes are compared by paired -test and McNemar test, respectively, and all results are significant with values <0.05 except the italic pairs. **The power of Wald test for Models A, B, and C were fixed as 0.6, whereas the average power of Wald test for Model D was 0.388 in average over all genes (left) and 0.377, 0.373, 0.365, 0.368, 0.354, 0.356, and 0.351 for , respectively, for genes with clustering results are different (right). †Bolded numbers are the maximum average power of MLC over different threshold values within each clustering method, trait model, and the set of genes (all or the ones with different clustering results by LDSelect and CLQ).