running GridLMM

[afrahshafquat]

I believe there have been several commits recently to this repo. Do you have a stable version of the software that I can use? I keep running into errors that get resolved in your later commits. Let me know if there is a stable version.

Right now I am getting this error:

[1] "Estimating h2_start via null model"
[1] "step_size: 0.333333333333333, non_zero: 0, num_to_test: 3"
Error in { : task 1 failed - "object 'n' not found"
Calls: GridLMM_GWAS -> GridLMM_posterior -> %do% ->
Execution halted

[afrahshafquat]

Also is there a test dataset to make sure that the program is actually running properly?

[deruncie]

Hi, Sorry about that. It should be fixed. Most of the commits have been bug fixes. I think the current version is as stable as any, but users keep running into issues. I hadn’t heard from anyone using the GridLMM_posterior function, so never noticed it was broken. I added a brief example of GridLMM_posterior to the end of the vignette. Hope this helps. Dan

…

On Nov 3, 2019, at 1:49 PM, afrahshafquat ***@***.***> wrote: I believe there have been several commits recently to this repo. Do you have a stable version of the software that I can use? I keep running into errors that get resolved in your later commits. Let me know if there is a stable version. Right now I am getting this error: [1] "Estimating h2_start via null model" [1] "step_size: 0.333333333333333, non_zero: 0, num_to_test: 3" Error in { : task 1 failed - "object 'n' not found" Calls: GridLMM_GWAS -> GridLMM_posterior -> %do% -> Execution halted — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#5?email_source=notifications&email_token=AB2IDTLSVSFJXG56ZXMW5RLQR42L7A5CNFSM4JINHXR2YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4HWPF2TQ>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AB2IDTPSBRAT7HWJWOJZXS3QR42L7ANCNFSM4JINHXRQ>.

[afrahshafquat]

Thanks Dan, I was able to run the program however, its giving me infinite values for Bayes Factor calculations. Could you help me fix the issue? Do i need to setup the prior differently here? Ideally I want to be able to run a standard linear mixed model GWAS and calculate Bayes Factor for each of the SNPs.

Thanks so much for all your help!

I am running the following command:
gwas = GridLMM_GWAS(formula=y~(1|ID),
test_formula = ~1,
reduced_formula = ~0,
data =data,
method='BF',
X = X-1,
inv_prior_X=0,
X_ID='ID',
relmat = list(ID=A_mat), verbose=T,max_steps = 100,h2_step = 0.01, centerX=T, scaleX=T, fillNAX=T,mc.cores = 1)

Here A_mat is a relatedness matrix, where for the final model: y = XB + u, where u ~ MVN(0, sigma*A_mat), where sigma is the variance parameter and B is the fixed effect vector for X (SNPs)

[deruncie]

Hi, Yes, `Inv_prior_X` is the precision of the Gaussian prior on the effect sizes. Setting this to zero means your prior is a Gaussian with infinite variance which is improper. Try setting this to 1 and see if you get finite values. But in GWAS, you expect small effects, so you might want to set this much larger. Since you have an intercept and a marker effect in you model, you could do Inv_prior_X = c(0,1), which would give a flat prior to the intercept and a N(0,1/1) prior to the marker effect. DAn

…

On Nov 9, 2019, at 10:34 AM, afrahshafquat ***@***.***> wrote: Thanks Dan, I was able to run the program however, its giving me infinite values for Bayes Factor calculations. Could you help me fix the issue? Do i need to setup the prior differently here? Ideally I want to be able to run a standard linear mixed model GWAS and calculate Bayes Factor for each of the SNPs. Thanks so much for all your help! I am running the following command: gwas = GridLMM_GWAS(formula=y~(1|ID), test_formula = ~1, reduced_formula = ~0, data =data, method='BF', X = X-1, inv_prior_X=0, X_ID='ID', relmat = list(ID=A_mat), verbose=T,max_steps = 100,h2_step = 0.01, centerX=T, scaleX=T, fillNAX=T,mc.cores = 1) — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#5?email_source=notifications&email_token=AB2IDTPBRC5MDMLN3WYEKOLQS37E5A5CNFSM4JINHXR2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEDUMJMQ#issuecomment-552125618>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AB2IDTOTZJLWBYWWSWZMSRLQS37E5ANCNFSM4JINHXRQ>.