simulating phenotypes for causal and non-causal SNPs separately

[fmoradi1365]

Hello,

I am using your PhenotypeSimulator package in my research and have a few questions:

I am interested in simulating phenotypes, based on an external genotype data, for both causal and non-causal SNPs to be able to compute type one error and power of some association tests. I followed the examples in the manual and vignette of the package but I did not understand how those R codes lead to simulating phenotypes separately for causal and non-causal SNPs. Do you mind sharing with me an example of R code that shows how traits of both causal and non-causal SNPs can be simulated based on a given genotype input matrix, if you have any?

Many thanks for your time.

[HannahVMeyer]

I am not entirely certain what your question is: genetic variant effects use SNP data provided by you or are simulated as independent, bi-allelic variants with defined frequency. Non-causal SNPs would not be included in the genetic variant effects, as they are not causal. Maybe you refer to genetic variants used for computing a genetic relationship matrix that builds the basis for the infinitesimal genetic effect?

Example 2 in this vignette shows the step-by-step simulation procedure and the combined approach via runSimulation: https://hannahvmeyer.github.io/PhenotypeSimulator/articles/PhenotypeSimulator.html

Please check these examples and if you still have question, specify in more detail what you mean by simulation of non-causal SNPs.

[HannahVMeyer]

I am closing this now, as this question might have also been clarified in #14 and #16. Feel free to re-open if question remains.

[fmoradi1365]

Hi Hannah,

Thanks for the reply and sorry for answering late.

Let me clarify my question: Let's say I am giving a genotype input file composed of 40 SNPs to PhenotypeSimulator and I set 5 of them to be considered as causal SNPs. PhenotypeSimulator generates a Phenotype matrix of N*P, in which N is the number of samples and P is the number of traits(phenotypes). Now I want to use this generated phenotype matrix to do a very popular multivariate association test. This test needs one genotype matrix and one phenotype matrix as inputs and outputs a set of p-values. When I consider those 5 causal SNPs as genotype file and the above-mentioned simulated phenotypes as phenotype file, I expect the majority of the 5 output p-values to be significant but they are not. Similarly, when I consider the remaining 35 noncausal SNPs as genotype file and the above-mentioned simulated phenotypes as phenotype file, the majority of the output p-values are unexpectedly significant. Did I miss understood usage of your package?

[HannahVMeyer]

Can you provide data were this happens? Are the genotypes you provide independent or correlated? I have shown in the paper that PhenotypeSimulator can use genome-wide data and only simulate causal effects for the the number of selected SNPs and variants in LD. As for the causal SNPs not be significant: how many samples did you simulate? how big was the beta you simulated? I cannot trouble-shoot any of this without more information i.e. at least your R code. Please provide and ideally also the genotype matrix (I will keep it confidential).

[fmoradi1365]

Many thanks for willing to help. Our genotypes are independent, and we have simulated 350 samples. We discussed the problem in our group again, and it seems that we have a solution. I will contact you in case we need more help.