ONLINE SUPPLEMENTARY MATERIAL FOR

Mapping Determinants of Human Gene Expression by Regional and Genome-

Wide Association

Vivian G. Cheung1,2,3, Richard S. Spielman2, Kathryn G. Ewens2, Teresa M.

Weber2,3, Michael Morley3, Joshua T. Burdick3

Online Table 1. Evidence of cis-association (with at least one SNP within

50 kb of target gene) for 374 phenotypes with variable evidence for cis-
linkage.

Phenotypes (65) with Phenotypes (133) with

Evidence of linkage
No. of Phenotypes evidence (P<0.001) of evidence (P<0.01) of
(t-values)
cis association# cis-association#

>5 27 19 (70%)** 22 (81%)**

4-4.9 33 13 (39%)** 17 (52%)

3-3.9 72 12 (17%)* 23 (32%)

2-2.9 242 21 (9%) 71 (29%)

# The proportion of phenotypes with cis association was compared to that among 50

randomly selected phenotypes by chi-square (** P < 1x10-4; * P < 0.05, df=1).
Power and Sample-Size

The power of the association analysis for one SNP was estimated in two ways.

First we carried out analyses on simulated data. We fixed total phenotypic variance (=1,

for convenience), and considered a range of values for R2. For each R2, we calculated

corresponding differences between genotype means for the three SNP genotypes

(assuming no dominance). Individual simulated phenotype values were obtained from

each of three normal distributions, in numbers drawn from Hardy-Weinberg proportions,

to reach the desired total sample size (60, 100, etc.). The process was repeated 500 times,

and standard linear regression analysis was done. The proportion of outcomes with P-

value <0.05/770,000 was taken as the power. Second, this procedure was checked by

comparing results with those from a published equivalent (Ball 2005).

For values between R2 = 0.4 and R2 = 0.5, the probability of detecting the effect

increases sharply, from 40% to >60% (in agreement with our findings using real data).

With samples of N=100, R2 of 0.25 or 0.35 corresponds to power of approximately 34%

and >80%, respectively. With samples of N=500, the power to detect R2 of 0.1 was > 80

%. Thus the sample sizes needed are large but nevertheless attainable for common

diseases.

Ball RD (2005) Experimental designs for reliable detection of linkage disequilibrium in

unstructured random population association studies. Genetics 170:859-73