Simulating multiple genetically correlated phenotypes

[fmorgante]

Hi Florian, thank you for a great package! I think it would be very useful if snp_simuPheno could be extended to simulate multiple genetically correlated phenotypes. I guess the framework could be made as flexible as one wants, but a great first step would be to allow for the effects to be drawn from a multivariate normal with mean 0 and a user-provided covariance matrix across traits, and independent residuals across traits.

Do you think this is something you are interested in implementing?

[privefl]

Yes, this is definitely something interesting to have.
E.g. see https://github.com/privefl/bigsnpr/blob/master/tmp-tests/test-gen-cor.R#L39-L53.

I guess it becomes a bit tricky if you consider that the two phenotypes can have different causal variants.

[fmorgante]

Yes, absolutely. I guess you could start by imposing the same causal variants across traits, but allowing an arbitrary number of traits. The user would then provide the desired genetic correlation matrix

[privefl]

Yes, that should be doable. Something like providing matrix of genetic correlations + vector of heritabilities might be easier than directly the covariance matrix.

[fmorgante]

That sounds great!

[privefl]

Do you want to try to implement it?

[fmorgante]

I do not have bandwidth at the moment, but I may be able to get to it later in the summer (if you or someone else does not do it before). I suggest we keep this issue open until someone implements this. What do you think?

[privefl]

Same. Sounds like a good idea.