Question: GENESIS: ascertainment in GWAS
0
4 months ago by
ferranpauls100 wrote:

Hi,

I'm using GENESIS package to do a GWAS and it works fine, but my dataset is based on families, which were ascertained by one individual each and I would like to know how to control this bias.

fitNullModel function allows you to adjust the model by a kinship matrix and a household membership matrix. I understand that I have to create a matrix of household membership with the probands but I'm not really sure, could someone help me, please?

modified 4 months ago by mconomos50 • written 4 months ago by ferranpauls100
0
4 months ago by
University of Washington
Stephanie M. Gogarten740 wrote:

Using a household matrix allows you adjust for the relationship between members of the same family who may not be genetically related (as the latter effect is captured in the kinship matrix), but share common environment, diet, etc. You can create a matrix in which cells for pairs of subjects in the same family are set to 1, and all other cells are set to 0 (with 1s on the diagonal). When you run fitNullModel, you would supply a list of both the kinship and household matrices: cov.mat=list(kinship=kin.mat, family=fam.mat)

Yes, but I think it's different of what I really want.

Ascertainment bias is adjusted by conditioning the likelihood of each family by the proband, which in my case is only only one individual.

If I do as you say, the variance of each family would be adjusted by all the probands at the same time, because you are interpreting them as a group instead of separating them by family.

No matter how I look at this, proband is not the same as household group. Is it possible to have a more detailed answer about this, please? I know I can build the household matrices, but it's not clear to me how I am adjusting by ascertainment by doing it...

0
4 months ago by
mconomos50
University of Washington, Seattle, WA, USA
mconomos50 wrote:

Hello,

I understand that your question revolves around accounting for the ascertainment bias from the sampling scheme of families; presumably an affected individual was identified for inclusion in the study and then family members were additionally included, thus leading to the bias. However, I am not sure how this is standardly accounted for in something like a mixed model. If you could further elaborate on how the model would be modified to to account for this, then we could help guide you as far as how to implement that analysis with GENESIS (assuming it's possible). This particular adjustment is not something that we have performed using GENESIS before.

Thanks