Snp - Quantitative Trait Association In A Gwa Study

Question

[Agatha]

I have a list of 700 snps and the associated genotypes for 8500 individuals. I am trying to analyze the association between the SNPs and two quantitative traits. ( => which snp is associated the most with each one of them).

I was considering to generate a linear regression model with each of the quantitative traits as Y variables, adjust by age and sex and then plot the log distribution of p values. (pvals for QT1) versus pvals for QT2).

regression QT1 : QT2, age, sex
regression QT2: QT1, age, sex

Is there a better approach/method/software tool in performing this type of analysis?

Thanks in advance.

Answer 1

[ff.cc.cc]

plink offers a fast approach to that kind of analysis with the parameters --qt-means and --GxE

[Agatha]

I have tried using plink for this , --assoc option , but I have only got the basic snp-qt association results. I have probably done something wrong when using --gxe returned only NA values...although I think I have used the correct input formats ..

Answer 2

[David]

Using regression (either logistic, linear, generalised linear model or generalised least squares) is the traditional method to perform such association analysis. It is efficient and robust. The secret is in controlling for confounding factor while not over doing it. I used R for in the past for doing that. It's fast enough for a research environment. You have to control for multiple hypothesis testing at the end of course and decide for an FDR threshold.

An additional step after running the bare bone approach above is to look for epistasis events. This is computationally intensive but there are several software out there. For example:

EIP: http://statgen.psu.edu/software/eip.html (R package)

FastEpistasis: http://www.ncbi.nlm.nih.gov/pubmed/20375113 (PLINK extension)

Answer 3

[yao.h.1988]

If there may be some confounding factor such as population structured or related individual in your sample , I would recommended that to use mixed Linear model to correct them , some papers in Nature (http://scholar.google.com.hk/scholar?q=mixed+model+GWAS&btnG=&hl=en&as_sdt=0%2C5) will introduced this method in GWAS

The other factor I think is important is that linear regression in GWAS often based on single marker regression,but we know SNPs are not independent, so it may loss something in this method.

Hope this help