Post by espride on Feb 5, 2014 21:40:47 GMT
We have recently been using GCTA to estimate the genetic variance across SNPs in two breeds of horses with recurrent exertional rhabdomyolysis from PLINK binary files. The Thoroughbred data (n=491 with 245 cases and 246 controls) was obtained on a 54K SNP chip and the Standardbred data (n=196 with 107 cases and 89 controls) were imputed from data from both the 54K and 74K SNP chips (total 72,000SNPs).
We have been running the data separately (TB and SB) and have been able to get good results when using the --reml (p-values 1.2e-6) on the GRM analyzed from all 31 autosomes and then pruned for individuals related >0.25 (these particular breeds of horses are more inbred and more stringent criteria leads to a small number of horses remaining in the analysis):
--autosome-num 31
--autosome
--maf 0.01
--grm-cutoff 0.25
Our next task was to partition the genetic variance onto individual chromosomes. We created GRM for each chromosome and used the remove option to exclude individuals that had been pruned from above (we also tried this without pruning individuals):
--autosome-num 31
--chr 1
--remove ind_list
--make-grm
I am able to run --reml for each individual chromosome but cannot get the analysis to work when using all 31 autosomes (--mgrm Multiple_GRM.txt). The analysis typically ends with an error that the log-likelihood would not converge (I have attempted --reml-maxit 10000 which did not correct the problem but consistently switched 12 to 13 files constrained). When trying smaller set of autosomes - I can get the analysis to run up to 10 chromosomes (1-10 or 2-11 or 10-20) but it always fails after the addition of an 11th file (typically smaller chromosomes). I have also tried to add the covariate of sex but have received the error the over half the files are constrained. I tried --reml-no-constrain but then received the error that the matrix was not invertible.
I have been able to get all 31 autosomes to work when using --reml-alg 2 but with a poor LRT value and insignificant p-values (0.45). Iam unsure on the validity of using the EM-reml algorithm versus the AI-reml or reml-alg 1 (which also does not converge) and was hoping for any suggestions you may have on running the analysis.
We have been running the data separately (TB and SB) and have been able to get good results when using the --reml (p-values 1.2e-6) on the GRM analyzed from all 31 autosomes and then pruned for individuals related >0.25 (these particular breeds of horses are more inbred and more stringent criteria leads to a small number of horses remaining in the analysis):
--autosome-num 31
--autosome
--maf 0.01
--grm-cutoff 0.25
Our next task was to partition the genetic variance onto individual chromosomes. We created GRM for each chromosome and used the remove option to exclude individuals that had been pruned from above (we also tried this without pruning individuals):
--autosome-num 31
--chr 1
--remove ind_list
--make-grm
I am able to run --reml for each individual chromosome but cannot get the analysis to work when using all 31 autosomes (--mgrm Multiple_GRM.txt). The analysis typically ends with an error that the log-likelihood would not converge (I have attempted --reml-maxit 10000 which did not correct the problem but consistently switched 12 to 13 files constrained). When trying smaller set of autosomes - I can get the analysis to run up to 10 chromosomes (1-10 or 2-11 or 10-20) but it always fails after the addition of an 11th file (typically smaller chromosomes). I have also tried to add the covariate of sex but have received the error the over half the files are constrained. I tried --reml-no-constrain but then received the error that the matrix was not invertible.
I have been able to get all 31 autosomes to work when using --reml-alg 2 but with a poor LRT value and insignificant p-values (0.45). Iam unsure on the validity of using the EM-reml algorithm versus the AI-reml or reml-alg 1 (which also does not converge) and was hoping for any suggestions you may have on running the analysis.