Run for longitudinal data

[ConnieXuhm]

Hi @deruncie !
I am working with a longitudinal data with 4 time points and want to explore the SNP*Time interaction. In lmer4, I run it with:

m<-lmer(SEVERE~Time+age+sex+SNP+SNP*time+(1+time|id),data=dat)

I want to achieve the same purpose on your GridLMM project, and add the GRM matrix. However, since I am weak in coding, I wonder how can I add the gxe into the formula? I would be so much appreciate if you could tell a little more details about the gxe interaction.
Thanks a lot!
Connie Xu

[ryshi06]

Hi Connie,

I am also doing longitudinal mixed models. I am wondering have you solved your problem? I also have another problem related to this. If there is no g*e interaction, I just want to add all the available time points into the model, how can I do this? The script is giving me an error like 'task 1 failed - 'names' attribute [1] must be the same length as the vector [0]".

Thanks for the help.
Ruyu

[deruncie]

Hi,

Connie, the syntax for a GxE test is:

gxe_gwas = GridLMM_GWAS(formula = SEVERE~Time+age+sex+(1 + time | id), 
                                                   test_formula = ~1 + time, # this is the full model ~SNP + SNP:time 
                                                   reduced_formula = ~1,  # this is the reduced model ~SNP
                                                   data = at,
                                                   ... plus the normal GridLMM arguments
                                                   )

or:

gxe_gwas = GridLMM_GWAS(formula = SEVERE~Time+age+sex+(0 + time | id), 
                                                   test_formula = ~0+ time,  # this is the full model ~SNP:time 
                                                   reduced_formula = ~1,     # this is the reduced model ~SNP
                                                   data = at,
                                                   ... plus the normal GridLMM arguments
                                                   )

The first fits an intercept for the SNP, and then a deviation for times 2-4. The second fits an separate SNP effect for each of the 4 times. They are approximately equivalent models, with the difference being that in the first case you're modeling a positive correlation of SNP (and id) effects across times, while in the second case you're modeling the 4 times as independent. Unfortunately right now GridLMM can't model the full correlations among the 4 times. For that you'd need to go to something like sommer or lme4qtl, but they'll be a lot slower.

Ruyu, I'd have to see move of your model to be able to help.

Dan

[ryshi06]

Hi Dan,

Thanks for the reply. I have initiated a new issue.

Ruyu

[ConnieXuhm]

Thanks @deruncie and @ryshi06 ,
I met the same error as issue#12, and doubt where the problem is. And also, I want to confirm that if GridLMM is able to deal with kinship matrix, since samples in my cohort are 1/3 related because they came from the same village. Thanks again!