|
Post by asling on Sept 4, 2015 11:56:57 GMT
Hello,
I am using option --reml-est-fix combined with --reml to estimate the variance explained by the SNPs as well as the fixed effect from the variable provided in --covar (sex) and --qcovar files (age). My trait is a binary trait (0->control, 1->case). The output file *hsq gave me the estimate of the fixed effect and the SE. However it is not clear for me that which direction on the outcome the fixed effect is. I saw both positive and negative values, but I am not sure how GCTA deals with binary variable (say, If a discrete variable with 2 possible value of 0 and 1 (provided in --covar) gets an positive estimate of fixed effect, does this mean that the level 1 of this variable will increase the chance to get disease? Or decrease? And how about that in a continuous variable provided in the --qcovar? )
Many thanks!
Best wishes,
Asling
|
|
|
Post by Jian Yang on Sept 21, 2015 11:32:23 GMT
|
|
|
Post by asling on Sept 21, 2015 20:56:14 GMT
Hello, Many thanks for your reply! Having checked the new post gcta.freeforums.net/thread/211/estimating-fixed-effects-gcta-greml, we found that there are still some mismatches between the new information and the old information that we got before. In out analysis we have 22 quantitative covariates (20PCs+age+age2)input by using --qcovar and 2 binary covariates(sex and another binary variable) input by using --covar. Having run --reml combined with --reml-est-fix, there are 25 lines of fixed effect (the lines starting below the line:Fix_eff SE) in the .hsq file. According to your new post, "If we have a quantitative covariate and a categorical covariate, the mean term will also be ignored", then we should have 26 lines of fixed effect estimated (22 lines for quantitative covariates and 4 lines(2X2) for binary covariates). Or if each of the binary covariates only takes one line, then we should have 24 lines in total, if so it seems that the first line should represent mean term(Although I don't understand how it is made)? Also in another thread gcta.freeforums.net/thread/129/adding-random-effect-reml we get a slightly different answer: Dec 11, 2014 12:54:50 GMT 1 zhihong said: Hi Asling, 1. The first row should be miu, population mean, 2. then quantitative covariates, 3. discrete covariates are at the bottom if you have. Cheers, Zhihong So I am not sure in the output *hsq file (1) if miu (mean) is listed in the first line and (2)if each binary covariate takes one line. Thanks, Asling
|
|
|
Post by Jian Yang on Sept 22, 2015 1:51:05 GMT
Sorry, it was my mistake. I've checked the source code again and have made a few changes in the post gcta.freeforums.net/thread/211/estimating-fixed-effects-gcta-greml. There are three key things 1. The mean term is always included. 2. The number of variables for a categorical covariate is t - 1 where t is the number of category. 3. For a categorial covariate, the effect of a category i is defined as the difference in mean phenotype between category i and category 1. In your case, the output will be mu + 20 PCs + age + age 2 + sex[2] + another_binary_variable[2]. So, there are 25 fixed effects in total.
|
|
|
Post by asling on Sept 22, 2015 8:51:44 GMT
Thanks, it is clear now.
Best wishes,
Asling
|
|