enisa
New Member
Posts: 2
|
Post by enisa on Sept 23, 2016 13:03:57 GMT
Hello,
I am using GCTA to estimate the dominant variance, but I get this error: the information matrix is not invertible. The sample is European, 1831 individuals.
These are the commands I use: 1) gcta64 --grm-gz CEUmatrix --keep idlist.txt --grm-cutoff 0.025 --make-grm-gz --thread-num 8 --out children 2) gcta64 --grm-gz children --keep children.grm.id --make-grm-d-gz --thread-num 8 --out children_d 3) echo "children">study.txt 4) echo "children_d">>study.txt 5) gcta64 --reml --mgrm-gz study.txt --pheno pheno.phen --covar sex.covar --qcovar ancestry.qcovar --thread-num 8 --out results --reml-no-lrt --reml-no-constrain > results.log
From the .log file, I get only one iteration:
Performing REML analysis ... (Note: may take hours depending on sample size). 2744 observations, 12 fixed effect(s), and 3 variance component(s)(including residual variance). Calculating prior values of variance components by EM-REML ... Updated prior values: 6.01661 6.01661 6.03699 logL: -5220.31 Running AI-REML algorithm ... Iter. logL V(G1) V(G2) V(e)
The input files are fine. Checked them several times, with a colleague also. So we have no clue now of what this error might be.. Tried to estimate the additive component and it works just fine!
I would really appreciate it if you could help! Thank you in advance!
|
|
|
Post by Zhihong Zhu on Sept 23, 2016 13:37:28 GMT
Hi enisa,
Dominance GRM is estimated from genotypes, e.g. gcta64 --bfile genotypes --make-grm-d --out grm_d.
Cheers, Zhihong
|
|
enisa
New Member
Posts: 2
|
Post by enisa on Sept 26, 2016 15:23:00 GMT
Hi Zhihong! Thank you very much for your answer! I tried it as you suggested, but I still get the same error. Could there perhaps be any other reason? That's what I get from the .log file now: Performing REML analysis ... (Note: may take hours depending on sample size). 1838 observations, 6 fixed effect(s), and 3 variance component(s)(including residual variance). Calculating prior values of variance components by EM-REML ... Updated prior values: 4.44262 4.44262 4.45223 logL: -3258.71 Running AI-REML algorithm ... Iter. logL V(G1) V(G2) V(e) 1 -3258.47 4.34505 3.38819 5.47437 2 -3257.90 3.03338 2.23193 7.62722 Thank you again! Enisa
|
|
|
Post by Zhihong Zhu on Sept 30, 2016 7:23:42 GMT
Hi Enisa,
"The information matrix is not invertible" suggests estimate of genetic variation might be negative. Because your additive variation is estimable, I guess that your dominance estimate might be extremely negative. Probably you could try only fitting dominance GRM and see the variance component, V(G1), in each iteration.
Cheers, Zhihong
|
|