1
$\begingroup$

I am trying to understand a GWAS paper which reports the effect of a SNP on outcome as a % variation, e.g., 15% variation in xxx levels.

The only information reported are the beta, standard error and p-values. I tried to read the paper but there is no mention of how this is calculated.

The paper in question is Genome-wide association meta-analysis for total serum bilirubin levels (Hum. Mol. Genetics, 2009).

Paragraph in question = These SNPs accounts for approximately 17.5, 18.1 and 16.7% of the variation...

If anyone has any clue, please help. I'll be most appreciative.

Improve this question
$\endgroup$
2
  • $\begingroup$ It looks to me like you can't calculate these values yourself from the data in the paper, because they don't give data on all four of the SNPs for each sample (there is only data for the pooled analysis in the supplementary data). But in general, I believe you could do this by calculating h^2 (heritability) from the regression model - although I have not done this combining multiple SNPs so I'm not sure how that might change the calculation. $\endgroup$
    – Ellie
    Commented Jan 6, 2014 at 16:41
  • $\begingroup$ Thanks for your input, I am trying to determine the % variation from one SNP in the supplementary data but it looks like I can't from whatever information provide. $\endgroup$
    – Sylvia
    Commented Jan 6, 2014 at 17:16

1 Answer 1

Reset to default
3
$\begingroup$

These appear to be simply $R^2$ values (also called coefficient of determination), that is, the variance accounted for in the outcome variable based on the covariates/predictors included in the model. In this case, the outcome variable is the bilirubin level (log transformed) and the covariates/predictors are the SNPs. In the footnotes of Table 1, the authors write: "The $\beta$ coefficient represents the change in log-transformed bilirubin with one additional copy of the allele modeled", so the authors used an additive model for the SNPs (where you use the number of alleles as the predictor).

In the FHS sample, the authors used a mixed-effects model to account for familial clustering/correlation. In such models, the computation of $R^2$ is a bit more intricate and, as far as I can tell, the authors do not provide a reference for what approach they used here. See, for example, this question for some references on this.

Improve this answer
$\endgroup$
1
  • $\begingroup$ Thank you for your comment, I will look at your links more thoroughly. $\endgroup$
    – Sylvia
    Commented Jan 6, 2014 at 17:14

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.