|
Post by carolina on Feb 5, 2016 9:34:02 GMT
Dear Yian... I have seen in many papers they exclude the cohort used to calculate LD (bfile in the file) from the meta-analysis but others then do not. Of course, I will prefer to keep all the cohorts in order not to compromise my power but not sure it is correct. Could you give me your opinion?.
In addition, I have in my meta-analysis some cohorts which are from admixed origin and have excluded them from the GCTA analysis as the LD is calculated in the CEU population, but from these cohorts some have less than 30% admixed individuals, and I am thinking perhaps I am being too stringent (particularly considering my tophits are quite common). As by doing so I lost 20% of the sample size in the meta-analysis and can not evaluate some of the loci GWS in the meta-analysis with all samples as they do not reach GWS in the only-CEU analysis.
|
|
|
Post by Jian Yang on Feb 9, 2016 6:36:45 GMT
"I have seen in many papers they exclude the cohort used to calculate LD (bfile in the file) from the meta-analysis but others then do not" I don't think so. In principle, the best reference sample is the combined sample of all the studies in the meta-analysis, which we usually don't have access to. We therefore could use one of the largest participating cohorts as the reference for LD. Please see gcta.freeforums.net/post/573/threadI'm not sure about that. If you can, you might add some admixed individuals to you reference sample for LD so that the proportion of admixed individuals is roughly the same as in the whole meta-analysis sample.
|
|
|
Post by carolina on Feb 9, 2016 10:06:25 GMT
Dear Jian, Thanks a lot for your answers. I have seen the forum already just was not clear about the need of excluding the cohort.
As for the admixed populations, not sure I understand can you elaborate a bit more. What is conflicting for me is the validity of the LD calculated. Lets say our population Generation R will have 50% of admixed individuals... but this admixture individually could vary some are mixtures or European with Africans, other with Asians other Latins... so how could including randomly some of these individuals help calculating an LD representative for the meta-analysis which includes cohorts of African Americans or Australian-admixed individuals.
It is just no clear. Thanks again. Carolina
|
|