QSNP and corresponding pvalues NA

[Merli Koitmäe]

Hi everyone,

I am running a 3-factor model with CFI=0.980:

F1 =~ T1+ T2 + T3 + T4 + T5 + T6 + T7

F2 =~ T1 + T8 + T9 + T10 + T5 + T7
F3=~ T1 + T2 + T9 + T10 + T4 + T5 + T6+ T7 + T11

T6 ~~ a*T6
a>0.001            

userGWAS(covstruc = LDSCoutput, SNPs = sumstats, model = model,
                 sub=c("F1~SNP", "F2~SNP", "F3~SNP"),smooth_check=TRUE,
                 fix_measurement=TRUE,Q_SNP=TRUE, parallel=T,cores=20, std.lv=TRUE, toler=1e-50)

There are no warnings when I run the usermodel and the GWAS output warning column has also zeros. However it produces NA-s in the QSNP column and hence the corresponding pvalue column. What could be causing this and how can I fix that?

Thanks in advance!

Merli

[agro...@gmail.com]

Hi Merli, 

I'm not able to reproduce this error on my end using a different multifactorial model. Could you post the full model syntax you ran? 

Thanks, 

  Andrew

[Merli Koitmäe]

Hi, Andrew!

Sure:

model <- 'F1 =~ NA*T2D + GDM + hba1c + fast_gluc + ins_sens + bmi + whr
             F2 =~ NA*T2D + T1D + Hyp + fast_ins + ins_sens + whr
             F3=~ NA*T2D + GDM + Hyp + hba1c + fast_ins + fast_gluc + ins_sens + bmi + whr + smoking
             bmi ~~ a*bmi
             a>0.001
     F1 ~ SNP
     F2 ~ SNP
     F3 ~ SNP
'  
CF<-userGWAS(covstruc = LDSCoutput, SNPs = dm_sumstats, model = model,

                 sub=c("F1~SNP", "F2~SNP", "F3~SNP"),smooth_check=TRUE,
                 fix_measurement=TRUE,Q_SNP=TRUE, parallel=T,cores=20, std.lv=TRUE, toler=1e-50)

Thanks for the help!

Merli

[agro...@gmail.com]

Hi Merli, 

Thanks for sending that along. I've tried running the same combination of arguments for my own three-factor model and still can't seem to replicate the error. I'm wondering if the issue has to do with the fact that you have white spacing between the factor and SNP name in the model syntax (e.g., F1 ~ SNP), but when specifying the sub argument do not have white spacing (F1~SNP). 

This issue came up earlier in the year when I first incorporated this new Q_SNP argument and should have been fixed so that you could specify the models and arguments as you have and still get Q_SNP values. See this previous google group posting: 

https://groups.google.com/g/genomic-sem-users/c/VoI2doy8ysY/m/3UOV-CtkAgAJ

Would you mind re-installing GenomicSEM, restarting R, and retrying this exact same code on a subset of SNPs to see if the issue persists when using the most up-to-date version of the package? 

If that still doesn't work, you could try removing the white spacing between the factor and SNP in the model syntax. I wouldn't think that's the issue as I'm able to get Q_SNP results when I use white spacing in my own model syntax and is something I had put in a code update for in response to the google posting referenced above, but still could be one thing to explore to rule out possible explanations for the Q_SNP NA values in the interest of troubleshooting. 

Best, 

  Andrew

[Merli Koitmäe]

Hi, Andrew!

Thanks for that!

Unfortunetaly I am running that in HPC cluster so I cannot update the GenomicSEM myself. I know they had some problems with installing in September, I do not know, what version they installed, but I asked them to update it just in case.

However, I tried running it without the white spacing, but still get NA-s for Q_SNPs :(

Merli

[Merli Koitmäe]

Dear Andrew,

They updated GenomicSEM in server, I reran the analysis on some SNPs, but the Q_SNP column is still all NA-s. Do you have any other ideas, what's going on and can I trust the other GWAS results?

Thanks!

Merli