Table 2 Comparison of weighted and unweighted GWAS results

From: Correcting for volunteer bias in GWAS increases SNP effect sizes and heritability estimates

 

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

Phenotype

r

\({N}_{eff}^{GWAS}\)

\({N}_{eff}^{WGWAS}\)

Increase S.E.s

sig. hits GWAS

sig. hits WGWAS

unique hits WGWAS

New loci

Age at First Birth

0.976 (0.013)

139,093

51,949

71%

30

3

2

0

BMI

0.992 (0.005)

372,969

135,238

77%

1205

127

5

0

Breast cancer

0.803*(0.038)

197,857

90,492

40%

45

8

4

1

Drinks per Week

0.936*(0.019)

265,696

96,008

83%

23

4

0

0

Self-rated health

0.973*(0.009)

372,714

136,982

82%

101

6

0

0

Height

0.993 (0.003)

374,175

151,328

61%

5114

1453

22

0

Physical activity

0.866*(0.031)

334,570

123,017

75%

3

0

0

0

Severe Obesity

0.949*(0.018)

373,834

136,396

75%

23

1

0

0

Type 1 Diabetes

0.660*(0.057)

373,786

132,605

87%

69

37

15

3

Years of Education

0.988 (0.006)

392,433

160,707

64%

331

49

3

0

  1. Comparisons use all UKB SNPs in HapMap3 (1,025,058 in total). The first column shows the genetic correlation r between GWAS and WGWAS results, estimated through LD-score regression (see “Methods”). The second and third columns show the effective sample sizes (see Methods) for GWAS and WGWAS. WGWAS increases standard errors by the percentage shown in column 4 (Increase S.E.s). column 5 shows the number of genome-wide significant SNPs for each trait in GWAS (P < 5 10−8, based on a two-sided t test); column 6 shows this in WGWAS; column 7 indicates how many of these genome-wide significant SNPs in WGWAS are unique. i.e., these SNPs have P < 5 10−8 in WGWAS, but P ≥ 5 10−8 in GWAS. Last, column 8 shows how many of these WGWAS-tagged new loci are unique hits, as indicated by a Hausman test that tests for genome-wide significance in the difference in the effect size as estimated through GWAS and WGWAS, a stringent test (see Methods). Thus, these loci were insignificant in GWAS, significant in WGWAS, and the difference in the effect sizes was genome-wide significant (PH < 5 × 10−8). * indicates values significantly different from one at a Bonferroni-corrected level of 5% significance, correcting for multiple hypothesis testing across ten phenotypes, i.e., (p < 0.05/10 = 0.005), based on a one-sided Z-test.