Selection of a "suggestive" level of association in a GWAS study
1
6
Entering edit mode
6.4 years ago
Vincent Laufer ★ 1.4k

In GWAS studies, 5*10-8 seems the defacto standard for association. A lesser known threshhold is 1*10-5.

This threshhold is often called "suggestive association."

I am writing to ask, where did this figure come from? Were empirical tests or statistical simulations done to select this level?

I can find many papers that USE this level, but far fewer that JUSTIFY it. Any relevant citations would be welcome, and thank you.

GWAS statistics association • 5.7k views
5
Entering edit mode
2.4 years ago
lindsay.leek ▴ 60

"The P values of Bonferroni corrected thresholds for suggestive, 5 and 1% genome-wide significant levels were 1, 0.05 and 0.01, respectively, divided by the number of SNPs used in the GWAS. The suggestive level was first proposed by Lander and Kruglyak [17] and represents the threshold where, under the null hypothesis, one false positive is expected per genome scan."

So the suggestive line is calculated as the -log10( 1 / number of variants)

Reference: Guo, Y., Huang, Y., Hou, L., Ma, J., Chen, C., Ai, H., ... & Ren, J. (2017). Genome-wide detection of genetic markers associated with growth and fatness in four pig populations using four approaches. Genetics Selection Evolution, 49(1), 21.

0
Entering edit mode

this is interesting - of course the bulk of the literature of which I am aware derives an understanding of these levels from the number of effectively independent markers.

so would be odd to use a pig based threshold unless they have similar levels of LD and numbers of haploblocks

1
Entering edit mode

Hi Vincent,

I added the paper because of " was first proposed by Lander and Kruglyak [17] " to indicate that they were the first to propose this. If you go to the paper of Lander Kruglyak, you will find a more precise explanation of the suggestive linkage. Good luck!

0
Entering edit mode

Thank you for clarifying, lindsay!