Question: Why Does A Rare Variant Need A Large Effect To Be Detected In Gwas?
gravatar for User 6659
8.2 years ago by
User 6659960
User 6659960 wrote:


I'm not from a biology background so please forgive the ignorance. I have just read that a rare variants (MAF < 0.5%) needs a large effect size to be detected by association studies. I don't follow the reasoning behind this. As long as the rare variant is on the DNA chip you are using to detect the genotypes of the individuals in the study, why should the rare variant need to have a larger effect than any other SNP on the chip. Is it just because the biological effect in question might be over-looked and not spotted at all as a biological effect if it is rare in the population? I don't really know what 'symptoms' are measured in these studies to know how it is possible to overlook an effect. In my limited conceptualisation the subjects in the studies either have or don't a detectable illness/trait but i imagine its more subtle than this.

thanks for your help

ADD COMMENTlink modified 8.2 years ago by Draco10 • written 8.2 years ago by User 6659960
gravatar for Lars Juhl Jensen
8.2 years ago by
Copenhagen, Denmark
Lars Juhl Jensen11k wrote:

It is simply a matter of statistics: if something has only a weak effect, you need a large number of observations for the effect to be statistically significant. If a SNP is rare, it will affect only a small proportion of your cohort. If at the same time it has only a weak effect, you would need a prohibitively large cohort to obtain a statistically significant result, especially after correction for multiple testing.

ADD COMMENTlink written 8.2 years ago by Lars Juhl Jensen11k

RE: the synthetic associations paper by Goldstein, PLoS Biology recently published two rebuttals, a rebuttal to the rebuttal by goldstein, and an editorial on the topic.

ADD REPLYlink written 8.2 years ago by Stephen2.7k

I appreciate this is a separate question. I can open a new thread but it didn't seem a big enough question. The paper is called Rare Variants Create Synthetic Genome Wide Assocations. I believe it caused quite a stir so I imagine many peopple will have read it

ADD REPLYlink written 8.2 years ago by User 6659960

Thank. I have another question if its ok. The paper i am reading says...'because recombination must be exceptionally high to eliminate the possibility of genome wide associations, one or more of the causal sites could be a considerable distance away from the common variant with the association signal'.. I don't get that. I thought the point was that there exists these regions called haplotypes where this is no recombination so you could use a tagSNP to tag other variants in the haplotype as they will be co-inherited. How can there be no recombination and 'exceptionally high recombination?

ADD REPLYlink written 8.2 years ago by User 6659960

I guess we are only missing the key word here: statistical power. The power to detect an association is both affected by effect size and population frequency.

ADD REPLYlink written 7.6 years ago by Adrian Cortes490
gravatar for Larry_Parnell
8.2 years ago by
Boston, MA USA
Larry_Parnell16k wrote:

I'd suggest exercising caution with regard to Goldstein's "synthetic association" theories. Yes, there have been some good examples of this, but in my opinion, which is shared by our group here, the rebuttals offer stronger evidence. Here's another.

More to the question as first posed - Lars is correct here.

Other reasons you may see weak effects, regardless of allele frequency, have to do with interactions. These can be gene-gene (epistasis), gene-environment, or gene-gene-environment. I just read an excellent overview of gene-environment interactions pertaining to asthma by Ober and Vercelli in Trends in Genetics. The examples described are worth noting for any association study.

ADD COMMENTlink modified 8.2 years ago • written 8.2 years ago by Larry_Parnell16k
gravatar for Draco
7.6 years ago by
Draco10 wrote:

Would you please give a link of your hypothesis: a rare/low-frequency variants (MAF < 5%) needs a large effect size to be detected by association studies.

I knew this is a common sense, but I failed to find any paper to support it. Though there are too many explained rare variants for association studies, difficult to find to cite.


ADD COMMENTlink modified 7.5 years ago • written 7.6 years ago by Draco10
Please log in to add an answer.


Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 2.3.0
Traffic: 2108 users visited in the last hour