Perform EVD outside relmatLmer/relmatGlmer

[variani]

In the current version 0.1.10, the relationship matrix is passed to relmatLmer and then it is decomposed by Cholesky/EVD. The decomposition operation is not visible to the user and there is little control over, e.g. errors. See also issue #1.

library(lme4qtl)
data(dat40)

mod <- relmatLmer(trait1 ~ 1 + (1|ID), dat40, relmat = list(ID = kin2))

It would be helpful to have something like this:

evd_kin2 <- eigen(kin2)

mod <- relmatLmer(trait1 ~ 1 + (1|ID), dat40, relmat = list(ID = evd_kin2))

[jfranco1951]

Dear Variani, Thanks for the lme4qtl, it is working well. I did try the EVD enhancement in a parallel long process but got an error message, then did try with the dat40 example and got the same message:

packageVersion("lme4qtl")
[1] ‘0.2.2’
evd_kin2 <- eigen(kin2)
mod <- relmatLmer(trait1 ~ 1 + (1|ID), dat40, relmat = list(ID = evd_kin2))
Error in relmat_lmer(trait1 ~ 1 + (1 | ID), dat40, relmat = list(ID = evd_kin2)) :
!is.null(rownames(relmat[[fn]])) is not TRUE

Thanks for help, J Franco

[variani]

Hi J Franco,

This issue is flagged as enhancement, meaning that I wish to implement this feature in the future. Though, I was not able to fine time for it, as you can see from the date the issue was created.

Currently, relmat argument supports only matrices with rownames to be IDs.

Best,
Andrey

[jfranco1951]

Dear Andrey, thank you for your answer, will look for another solution, maybe the matml?, J Franco --------------------- Jorge Franco D. Profesor Libre, Departamento de Biometría, Estadística y Computación Facultad de Agronomía, UDELAR Ruta 3, Km. 363, Paysandú URUGUAY telefono: +598 99123946

…

On Mon, Jun 14, 2021 at 9:08 AM Andrey Ziyatdinov ***@***.***> wrote: Hi J Franco, This issue is flagged as enhancement, meaning that I wish to implement this feature in the future. Though, I was not able to fine time for it, as you can see from the date the issue was created. Currently, relmat argument supports only matrices with rownames to be IDs. Best, Andrey — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#6 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AGKI4M6BRCE7LE73P7W4W63TSYEPFANCNFSM4EULVRFA> .

[variani]

You can try matlm package if it meets your needs. I should say that matlm is experimental and I mainly developed it to implement custom models for my projects. It has documentation and examples, but it is even less official than lme4qtl.

[jfranco1951]

Thanks for the tip! J Franco

On Thu 17 Jun 2021 at 15:53 Andrey Ziyatdinov ***@***.***> wrote: You can try matlm <https://github.com/variani/matlm> package if it meets your needs. I should say that matlm is experimental and I mainly developed it to implement custom models for my projects. It has documentation and examples, but it is even less official than lme4qtl. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#6 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AGKI4M4L533G6ZMW3HVC2N3TTJACJANCNFSM4EULVRFA> .

-- --------------------- Jorge Franco D. Profesor Libre, Departamento de Biometría, Estadística y Computación Facultad de Agronomía, UDELAR Ruta 3, Km. 363, Paysandú URUGUAY telefono: +598 99123946

[igorkf]

Hello @variani . Thanks for taking the time to write this package.
Regarding the EVD implementation, the BGLR can pass the eigendecomposition object to the model instead of the kinship matrix: https://github.com/gdlc/BGLR-R. For example, you can see here: https://github.com/gdlc/BGLR-R/blob/master/inst/md/GBLUP.md#RKHS2
I don't know how their implementation was, but it might be helpful if you are still thinking of implementing this functionality.