Re: Low Neff and mean chi^2 for latent factor in a hierarchical model

[Mingyi Du]

For additional information:

My Factor 2 contains Variable 4, which is HDL. I applied reverse-coding for HDL sumstats since SNP effects in HDL GWAS are associated with increased HDL levels, which are beneficial for my second-order outcome. I applied reverse-coding to align the effect direction. The subsequent ldsc output and GWAS for multivariate GWAS analysis are constructed based on revserse-coded HDL sumstats. I wonder if the current issue relates to this process. 

Best,

Mingyi Du

在2025年2月12日星期三 UTC+8 19:19:18<Mingyi Du> 写道：

Hi all,
First of all, thank you for the excellent package and this is really great work.
I just have a few questions:
I am running a hierarchical model using the userGWAS() function, which looks like this:

# hierarchical model
model <- "F1 =~ 1*Variable1 + Variable2 + Variable3
        F2 =~ 1*Variable4 + Variable5 + Variable6
        F3 =~ 1*Variable7 + Variable8
        OutcomeFactor =~ 1*F1 + F2 + F3
        F1 ~~ 0*F2
        F1 ~~ 0*F3
        F2 ~~ 0*F3
        Variable6 ~~ a*Variable6
        Variable2 ~~ b*Variable2
        Variable1 ~~ c*Variable1
        Variable4 ~~ d*Variable4
        Variable5 ~~ e*Variable5
        Variable8 ~~ f*Variable8
        Variable7 ~~ g*Variable7
        F1 ~~ i*F1
        F2 ~~ j*F2
        F3 ~~ k*F3
        OutcomeFactor ~~ l*OutcomeFactor
        a > 0.001
        b > 0.001
        c > 0.001
        d > 0.001
        e > 0.001
        f > 0.001
        g > 0.001
        h > 0.001
        i > 0.001
        j > 0.001
        k > 0.001
        l > 0.001
        OutcomeFactor ~ SNP
"
After Obtaining the GWAS for OutcomeFactor which is a second-order factor, I calculate the Neff (N=1,568,098) using equations listed on https://github.com/GenomicSEM/GenomicSEM/wiki/5.-Multivariate-GWAS.

After this, I plan to obtain GWAS summary statistics for all my first-order factors (F1, F2, F3) using code below:

#first-order GWAS
model <- "F1 =~ 1*Variable1 + Variable2 + Variable3
        F2 =~ 1*Variable4 + Variable5 + Variable6
        F3 =~ 1*Variable7 + Variable8
        OutcomeFactor =~ 1*F1 + F2 + F3
        F1 ~~ 0*F2
        F1 ~~ 0*F3
        F2 ~~ 0*F3
        Variable6 ~~ a*Variable6
        Variable2 ~~ b*Variable2
        Variable1 ~~ c*Variable1
        Variable4 ~~ d*Variable4
        Variable5 ~~ e*Variable5
        Variable8 ~~ f*Variable8
        Variable7 ~~ g*Variable7
        F1 ~~ i*F1
        F2 ~~ j*F2
        F3 ~~ k*F3
        OutcomeFactor ~~ l*OutcomeFactor
        a > 0.001
        b > 0.001
        c > 0.001
        d > 0.001
        e > 0.001
        f > 0.001
        g > 0.001
        h > 0.001
        i > 0.001
        j > 0.001
        k > 0.001
        l > 0.001
        OutcomeFactor ~ 0*SNP
        F1 ~~ 0*SNP
         F2 ~~ 0*SNP
         F3 ~~ 0*SNP
         F1 ~ SNP
         F2 ~ SNP
          F3 ~ SNP  
"

MASLD.GWAS <- userGWAS(covstruc=MASLD_ldscout,
                       SNPs=MASLD_sumstat,
                       estimation="DWLS",
                       model=model,
                       cores=16,
                       toler=1e-50,
                       SNPSE=FALSE,
                       sub=c('F1 ~ SNP', 'F2 ~ SNP', 'F3 ~ SNP'),
                       parallel=TRUE,
                       GC="standard",
                       MPI=FALSE,
                       smooth_check=FALSE,
                       printwarn=TRUE)



This ouputs all F1 ~ SNP","F2 ~ SNP", "F3 ~ SNP" results, and I select columns only contain "F1 ~ SNP" as my F1 GWAS, etc.
First of all, I wonder if this processing of F1, F2, F3 GWAS  is right.  
Second, when I use the same method for calculating Neff for my F1, F2, and F3. However, the Neff for F2 is extremely low, compared to F1 and F2.
Neff for F1: 1,568,098; 

F2: 196,429;

 F3: 1,060,424

Morever, I used the F2_GWAS summary statistics to perform LDSC and the results indicated low mean Chi^2:
Total Observed scale h2: 0.0566 (0.0059)
Lambda GC: 0.867
Mean Chi^2: 0.9508
Intercept: 0.7317 (0.007)
Ratio: NA (mean chi^2 < 1)

Everything goes reasonable for F1, F3, and my second-order OutcomeFactor. I struggled to find out what is wrong here. I would really appreciate if anyone could give me some advices.

Thanks very much!

[Michel Nivard]

Hi Mingyi,

Can you provide the model fi/results for the factor model without the SNP for us? What happens if you dont reverse code?

Best,

Michel

--
You received this message because you are subscribed to the Google Groups "Genomic SEM Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to genomic-sem-us...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/genomic-sem-users/21b3aa44-8c0c-41b6-bad4-ab19604e8228n%40googlegroups.com.