Perform joint analysis scan

now1
New Member

Posts: 2

Perform joint analysis scan Jun 20, 2016 16:03:29 GMT

Quote

Post by now1 on Jun 20, 2016 16:03:29 GMT

Hi

Both SNPTEST and plink have a "condition" flag that enables the user to perform a scan where every snp is conditioned on a given snp or list of snps (rs1..rsn). The per snp (say for snp x) results that are given back correspond to fitting a joint model of pheno~rs1+rs2+..rsn+snpx.

Is there a way of getting results for this same joint model in GCTA applied to case-control summary statistics other than by repeated calls to gcta with --extract <rs1...rsn>,snpx --cojo-joint? I've looked into the --cojo-cond flag but based on the paper ("Conditional and joint multiple-SNP analysis...") I think this estimates results (bC etc) based on regressing the residuals of pheno~rs1+rs1+rsn against snpx which is somewhat different.

Thanks,
Nick

Jian Yang Administrator Posts: 362	Perform joint analysis scan Jun 21, 2016 0:28:02 GMT Quote Select Post Deselect Post Link to Post Member Give Gift Back to Top Post by Jian Yang on Jun 21, 2016 0:28:02 GMT --cojo-cond is fitting a model of pheno~rs1+rs2+..rsn+snpx using summary data.

now1
New Member

Posts: 2

Perform joint analysis scan Jun 21, 2016 10:05:44 GMT

Quote

Post by now1 on Jun 21, 2016 10:05:44 GMT

Hi Jiang

Many thanks for the quick response.

I was prompted to ask the question because when I run the joint analysis and the conditional analysis against the same snps (say rs1, rs2 and snpx) and the same .ma file I get different results. I've now looked more closely and see that I do indeed get exactly the same p-value (pC vs pJ) for snpx but that the effect sizes bC and bJ seem to be estimated on a different scale (the effect sizes and corresponding standard error differ by a factor ~1.6 ). In both cases the reported estimated phenotypic variance is the same and the estimated sample size for snpx is also the same. I'm using gcta 1.25.3.

Thanks again,
Nick

Jian Yang
Administrator

Posts: 362

Perform joint analysis scan Jun 22, 2016 2:50:03 GMT

Quote

Post by Jian Yang on Jun 22, 2016 2:50:03 GMT

Hi Nick,

They are different.
Conditional analysis ~= adjusting the phenotype by the covariate SNPs and then testing association between the target SNP and the adjusted phenotype.
Joint analysis fits all the SNPs in the model ~= adjusting both the phenotype and the target SNP by the covariate SNPs and then testing association between the adjusted target SNP and the adjusted phenotype.

You could check this in R.

Cheers,
Jian

Last Edit: Jun 22, 2016 2:50:44 GMT by Jian Yang

Post by now1 on Jun 20, 2016 16:03:29 GMT

Post by Jian Yang on Jun 21, 2016 0:28:02 GMT

Post by now1 on Jun 21, 2016 10:05:44 GMT

Post by Jian Yang on Jun 22, 2016 2:50:03 GMT