Post by dhui on Sept 12, 2023 21:19:06 GMT
Hi,
Edit: Ran a small test and the results seem to be identical, just wanted to double check this is right?
To run FastGWA faster, can I run it per chromosome instead of all chromosomes per once? For instance, in the --mbfile file, can I just have a separate file per chromosome and run 22 jobs, instead of putting all 22 chromosomes in --mbfile? The estimated running time for ~6500 individuals and 30M SNPs (all chromosomes) is 14329.4 minutes (~10 days).
I had a couple other questions as well. There is no phenotype transformation done for FastGWA, correct? So if I wish to rank-normal transform my phenotype I should do this myself before running the analysis. Also, I saw this on the FastGWA page: "Note: when there are covariates fitted in the model (by the --covar or --qcovar flag), to accelerate the computation, we regress out the covariates from the phenotype and use the adjusted phenotype for the fastGWA or linear regression analysis". Is it possible to turn this off, as it does on the MLMA page (parameter "--mlma-no-preadj-covar")? Also, is the random genetic effect in FastGWA leave-one-chromosome-out, as there seems to be an option for MLMA to do so (--mlma-loco)?
I was thinking of just using MLMA instead of FastGWA for our analysis, as the max sample size for our different phenotypes is only ~6,500. But I was also hoping to see the sample size per SNP in the output, which doesn't seem to be in the MLMA output ("chromosome, SNP, physical position, reference allele (the coded effect allele), the other allele, frequency of the reference allele, SNP effect, standard error and p-value"). We combined WGS and array samples together, and removed SNPs that were imputed poorly from the array samples, so some SNPs have much lower N than others. Curious if there's any recommendation here?
Thanks.
Daniel
Edit: Ran a small test and the results seem to be identical, just wanted to double check this is right?
To run FastGWA faster, can I run it per chromosome instead of all chromosomes per once? For instance, in the --mbfile file, can I just have a separate file per chromosome and run 22 jobs, instead of putting all 22 chromosomes in --mbfile? The estimated running time for ~6500 individuals and 30M SNPs (all chromosomes) is 14329.4 minutes (~10 days).
I had a couple other questions as well. There is no phenotype transformation done for FastGWA, correct? So if I wish to rank-normal transform my phenotype I should do this myself before running the analysis. Also, I saw this on the FastGWA page: "Note: when there are covariates fitted in the model (by the --covar or --qcovar flag), to accelerate the computation, we regress out the covariates from the phenotype and use the adjusted phenotype for the fastGWA or linear regression analysis". Is it possible to turn this off, as it does on the MLMA page (parameter "--mlma-no-preadj-covar")? Also, is the random genetic effect in FastGWA leave-one-chromosome-out, as there seems to be an option for MLMA to do so (--mlma-loco)?
I was thinking of just using MLMA instead of FastGWA for our analysis, as the max sample size for our different phenotypes is only ~6,500. But I was also hoping to see the sample size per SNP in the output, which doesn't seem to be in the MLMA output ("chromosome, SNP, physical position, reference allele (the coded effect allele), the other allele, frequency of the reference allele, SNP effect, standard error and p-value"). We combined WGS and array samples together, and removed SNPs that were imputed poorly from the array samples, so some SNPs have much lower N than others. Curious if there's any recommendation here?
Thanks.
Daniel