### Post by Jian Yang on Sept 21, 2015 1:42:06 GMT

If there is only one genetic variance component (i.e. a single GRM) in your analysis, GCTA will calculate the LRT for the genetic variance automatically. The log likelihood for the full model (logL) and that for the reduced model (logL0) as well as the LRT and p-value will be reported in the *.hsq file, where LRT = 2[logL - logL0] which is distributed as a mixture of 0 and chi-squared (df = 1) with a probability of 0.5.

If you have multiple genetic variance components involved in your analysis (e.g. an analysis of genotype-environment (GE) interaction or a joint analysis of all chromosomes), by default, GCTA will only provide the LRT for first genetic variance component. In this case, you may need use the option --reml-lrt to specify which component(s) you want to test. For example, for a GE interaction model, y = Xb + e + g + ge + e, if you want to test the significance of the variance of GE interaction effects, you can add the option --reml-lrt 2 to your REML analysis:

You can also calculate the LRT for multiple genetic variance components. For example, for a joint analysis of 22 chromosomes (22 genetic components in the model), you could test whether, for example, chromosomes 3 and 7 simultaneously by adding the option --reml-lrt 3 7 to the analysis:

The LRT for multiple components is distributed as a mixture of 0 and chi-squared (df = p) with a probability of 0.5, where p is the number of components to be tested.

If you have multiple genetic variance components involved in your analysis (e.g. an analysis of genotype-environment (GE) interaction or a joint analysis of all chromosomes), by default, GCTA will only provide the LRT for first genetic variance component. In this case, you may need use the option --reml-lrt to specify which component(s) you want to test. For example, for a GE interaction model, y = Xb + e + g + ge + e, if you want to test the significance of the variance of GE interaction effects, you can add the option --reml-lrt 2 to your REML analysis:

`gcta64 --grm test --pheno test.phen --gxe test.gxe --reml --reml-lrt 2--out test`

You can also calculate the LRT for multiple genetic variance components. For example, for a joint analysis of 22 chromosomes (22 genetic components in the model), you could test whether, for example, chromosomes 3 and 7 simultaneously by adding the option --reml-lrt 3 7 to the analysis:

`gcta64 --mgrm grm_chrs.txt --pheno test.phen --reml --reml-lrt 3 7 --out test_chrs`

The LRT for multiple components is distributed as a mixture of 0 and chi-squared (df = p) with a probability of 0.5, where p is the number of components to be tested.