snowy
New Member
Posts: 1
|
Post by snowy on Jan 29, 2016 1:25:43 GMT
Hello, everyone
I’m starting learning GWAS and I have used --mlma to do association analyses for one quantitative trait based on two populations. Although all the processes worked well, I have several questions.
(1) In my first round, error is that Log-likelihood not converged, it continued to yield these errors until I added command “--reml-maxit 1”. So when there is no iteration, are the results reliable? and what caused the errors with iterations?
(2) In output file, the column ‘ b’ denotes SNP effects and ‘p’ column represents p-value. Then, how do I decide the significant association SNPs with my trait, by the largest SNP effect or by the smallest P-value?
(3) I used 10 PCs as covariates to correct for population stratification initially. But I found that you suggested no PCs and thought GRM was enough to correct bias in previous Posts. I conducted these two scenarios and got different results. So I don’t know which one is more reliable.
(4) Some samples in my data have relatedness, but if I exclude them, there are negative impact on samples size. Should I remove these related individuals?
Thank you!
|
|
|
Post by Jian Yang on Jan 29, 2016 2:05:19 GMT
Re 1) Yes, it's not reliable. You might try --reml-alg 1. If it still doesn't work, I would suggest you use linear regression analysis with PCs (e.g. --linear option in PLINK). You might remove related sample using GCTA --grm-cutoff 0.05 because fitting PCs does not account for relatedness.
Re 2) Smaller p-value means more significant.
Re 3) You don't need to fit PCs if you have fitted the whole GRM. Fitting a GRM is equivalent to fitting all PCs.
Re 4) A MLMA analysis would account for relatedness.
|
|