University of Michigan Center for Statistical 
Genetics
Search
 
 

 
 

Part 2: Power for a candidate gene study

You may be curious as to the potential benefit of using a two-stage design to reduce costs when examining far fewer markers than those interrogated during a genome-wide association study, for example, when conducting a candidate gene study.

We will address this question by considering the following scenario. Suppose that you have a collection of 1000 cases and 1000 controls, and would like to test 2000 SNPs located in and around candidate genes. Assume again that the SNPs being tested are independent and that you are willing to tolerate 2 false positives, allowing the use of a false postive rate of 2/2000 = .001. We can begin a power analysis as we did before, by determining what size effects you will be able detect with high power using a one-stage design. Again, we will consider the case when the disease of interests has prevalence of 10% and the risk allele that has frequency of .30 in the population. Consider the results below.

Screenshot

The above screen shot shows that a disease variant with GRR of 1.32 can be detected with 90% power using a one-stage design. And, as with the genome-wide association study, we see that the same power can be achieved when genotyping 50% of the sample in stage 1 and following up 50% of the markers into the second stage. We can again explore how genotyping can be reduced while still retaining much of the one-stage power. By playing with sliders controlling the percentage of samples genotyped in Stage 1 and the percentage of markers genotyped in stage 2, we find that we can reduce the amount of genotyping by 49% while still achieving power of about 88% by genotyping 35% of the sample in stage 1 and selecting 24% of the markers for follow-up in stage 2.

You can continue the tutorial by learning about optimal designs or you can return to the main tutorial menu and pick a different topic.


 
 

University of Michigan | School of Public Health | Abecasis Lab