1
$\begingroup$

I did a GWAS analysis in the past for antibiotic resistance of E. Coli and the results were interesting (matching the literature). I did a new GWAS analysis for some new samples, but the results are not in agreement with the previous GWAS and the literature. So, I was wondering if the combination of them would be in agreement with the literature, as having more samples to support an argument is always a good idea.

So, my question is how I can combine two GWAS study and what quality control steps do I need to follow?

$\endgroup$
1
  • 1
    $\begingroup$ Hi there, you got to be much more specific with your questions on this webpage. You should write the reasoning behind combining the combining and what do you want to do with the combined GWAS output. What did you consider already, what you found when you tried to research this question on your own... You can edit your question and add all that :-) $\endgroup$
    – Kamil S Jaron
    May 5 '20 at 13:59
0
$\begingroup$

The first way is to combine two samples and re-run GWAS. A second way is to do a meta-analysis using GWAS results.

$\endgroup$
5
  • $\begingroup$ For combining them and re-running GWAS, how should I take into account batch-effect, as these groups belong to two different batches? What do you mean by meta-analysis exactly? $\endgroup$
    – Negin
    May 5 '20 at 16:10
  • 1
    $\begingroup$ Batches may be a problem. I do not know how to do it in E. coli, but if it's human, I would do a PCA to see if the two batches overlap. I think the batch effect is moderate in genotype data. In meta-analysis, you can just use the effect sizes and p-values from many studies to calculate a new p-value. You can read some paper in human GWAS perhaps. $\endgroup$
    – Phoenix Mu
    May 5 '20 at 18:13
  • $\begingroup$ What about SNP missingness for different batches? Should I remove the SNPs with high missingness for one batch and low missingness for another batch? Or Should I look at the average missingness. $\endgroup$
    – Negin
    May 6 '20 at 10:11
  • $\begingroup$ Any answer, dear @Phoenix Mu? $\endgroup$
    – Negin
    May 6 '20 at 14:48
  • $\begingroup$ I think if you want to merge the genotypes and re-do GWAS, you should look at overall (average) missingness. I have no idea about the SNP missingness in E. Coli. In human, it is usually a very small proportion of SNPs that are missing and should not be a concern. $\endgroup$
    – Phoenix Mu
    May 6 '20 at 17:37
0
$\begingroup$

I'd recommend first having a think about how you would go about combining two GWAS with the same sample set (e.g. split the samples randomly in two). Even with the same source population, you'll get different associations coming up by sampling a different sub-set of that population.

Once you can do that multiple times, you can attach a level of confidence to each SNP association within each study, which can be treated as a prior probability for the purpose of combining multiple independent studies.

It sounds like you're in a great position to be able to do this, because you've done the analysis on both groups and [presumably] have all the genotype-level data from both studies.

$\endgroup$
4
  • $\begingroup$ Dear @gringer I am not sure if I understood you totally. First, it is good to know that my samples in the two groups are different. Do you mean I don't stick with these two groups, rather I mix them, iteratively?! $\endgroup$
    – Negin
    May 6 '20 at 14:47
  • $\begingroup$ First, stick within a single group. That way you don't need to worry about batch effects and SNP dropout. How would you compare two different GWAS taken from the same group? $\endgroup$
    – gringer
    May 6 '20 at 20:59
  • $\begingroup$ But, my two groups are coming from different sources, that's why I didn't mix them from beginning. $\endgroup$
    – Negin
    May 6 '20 at 21:20
  • $\begingroup$ Yes, I'm aware of that. But it's a good idea to start with something that works, and extend it to do what you want it to do. Work out first how you'd combine results from the same source. If you can't do that, there's no hope in combining results from different sources. Once that's sorted out, see how much of that combination process can be applied to results from different sources. $\endgroup$
    – gringer
    May 7 '20 at 2:08

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