|
Post by ks on Apr 7, 2014 14:29:07 GMT
Hi! I would like to ask two questions.
First, I got the message "Error: No SNP is retained for analysis", when I tried to make GRM. I used a list for extraction including 100 SNPs. I guess the number of SNPs so little that I got the massage. Is it correct ??
Second, I got below result after REML calculation. SE and p-value were surely small, but p-value was 0. Does this mean that p-value was very very small or the calculation was not completed correctly? If the p-value was 0 because p-value was very small, can I get the correct value like 1.2345 x E-12345 ?
Source Variance SE
V(G) 88.524256 5.723454
V(e) 93.175200 2.534392
Vp 181.699457 6.152210
V(G)/Vp 0.487202 0.017809
logL -9669.104
logL0 -11212.040
LRT 3085.872
df 1
Pval 0
n 3294
|
|
|
Post by Jian Yang on Apr 8, 2014 5:24:41 GMT
Re 1) No, in principle, you can use any number of SNPs. Re 2) Yes, LRT is distributed with 0.5 probability of 0 and 0.5 probability of chi-squared (df = 1).
|
|
|
Post by ks on Apr 8, 2014 7:39:36 GMT
Dr. Jian
Thank you for your response.
I'm going to try find the cause about Question 1. I will doubt all SNPs are excluded under the condition I used in the
About Question 2, I can understand. However I have not seen usually such a large chi-squared value. Can this result be reliable, if data set doesn't contain wrong ??
ks
|
|
|
Post by Jian Yang on Apr 8, 2014 10:52:40 GMT
Re 1) It should just be a format issue. Re 2) My guess is that you used a subset of SNPs rather than all the SNPs. Am I right? If the SNPs were selected from GWAS using the same samples, then the estimate will biased upwards.
|
|
|
Post by ks on Apr 8, 2014 11:59:46 GMT
Dr. Jian
Thank you again for your quick response.
Yes, surely I used a subset of SNPs.
First, I used all SNPs using in the imputation for a SNP-list. However, SE and p-value were large.
Reading topics in this board, I learned small SE was important for correct analyses, and if the number of SNPs was large,and/or sample size was too small, SE tend to be large one.
I think I read a paper suggesting "you need at least 3000 samples for GCTA analyses". Because I used more than 3000 samples and I could not increase samples, I tried to select the SNPs using various ways. The result I paste to this board was one of them, and as you appointed, I selected from GWAS data using the same samples.
Q1) Can't I adjusted the bias by some ways using SNP-list arbitrarily selected by GWAS and/or other criteria? Q2) When the answer is "No I cannot use such list", is it difficult to use GCTA if I cannot increase the sample numbers?
|
|
|
Post by Jian Yang on Apr 8, 2014 14:32:19 GMT
Re Q1) If the SNPs are not selected based on association p-value or other relevant statistics, it should be fine. Re Q2) You still can and the SE is expected to be ~0.1 which is not too bad if the estimate is large.
|
|
|
Post by ks on Apr 9, 2014 1:48:04 GMT
Thank you very much, Dr Jian.
I will learn and try again following your kind advises.
ks
|
|