GWAS-by-subtraction inquiry: extraordinary high outputGWAS() p-value

[laogou]

Hi, I'm Weichen Song from Shanghai Jiaotong University. I tried to use your excellent tool GWAS-by-subtraction and the result from LDSC and usermodel() seems reasonable (see below). But the userGWAS returned very high p-value: the lowest p-value for >5m SNPs was 0.05, which definitely violates a uniform distribution:

head(d[order(d[,15]),])
SNP CHR BP MAF A1 A2 lhs op rhs free label
57837338 rs836952 9 25904637 0.0685885 C T ncm ~ SNP 5
57709235 rs79300279 9 6133663 0.0407555 C A ncm ~ SNP 5
57837336 rs1219935 9 25902763 0.0755467 C T ncm ~ SNP 5
5889791 rs11198504 10 120368534 0.0536779 T G ncm ~ SNP 5
58897836 rs11198501 10 120367850 0.0536779 G C ncm ~ SNP 5
512627839 rs3865368 18 11974749 0.0695825 T G ncm ~ SNP 5
est SE Z_Estimate Pval_Estimate chisq
57837338 -0.1268818 0.06668243 -1.902776 0.05706972 3.432989e-17
57709235 -0.1427829 0.07660701 -1.863835 0.06234482 1.949406e-16
57837336 -0.1174886 0.06424197 -1.828846 0.06742272 1.225387e-15
5889791 -0.1270428 0.07218130 -1.760052 0.07839900 9.414978e-18
58897836 -0.1269866 0.07218207 -1.759254 0.07853435 2.210379e-16
512627839 -0.1137829 0.06561039 -1.734221 0.08287881 2.639530e-16
chisq_df chisq_pval AIC error warning
57837338 0 0 12 0 0
57709235 0 0 12 0 0
57837336 0 0 12 0 0
5889791 0 0 12 0 0
58897836 0 0 12 0 0
512627839 0 0 12 0 0

 It is acceptable that the latent factor ncm does not have a significant heritability, but i suppose such high p-value cannot be explained by random effect. Could you provide some suggestion on the issue? Thank you very much for your help.

My usermodel() model and outcome:

model<-'cm=~NA*MDD + start(0.1)*CM

        ncm=~NA*MDD

         

         ncm~~1*ncm

         cm~~1*cm

         cm~~0*ncm

         CM ~~ 0*MDD

         CM~~0*CM

         MDD~~0*MDD'

output
$modelfit
chisq df p_chisq AIC CFI SRMR
df NA 0 NA NA NA NA
$results
lhs op rhs Unstand_Est Unstand_SE STD_Genotype STD_Genotype_SE
5 cm =~ MDD 0.1089702 0.00816797489531124 0.6697152 0.0501992078896497
4 cm =~ CM 0.2379085 0.0101221997048112 1.0000000 0.0425466059304789
9 ncm =~ MDD 0.1208323 0.00812108043689905 0.7426180 0.0499110012486965
8 ncm ~~ ncm 1.0000000 1.0000000
1 cm ~~ cm 1.0000000 1.0000000
3 cm ~~ ncm 0.0000000 0.0000000
6 MDD ~~ CM 0.0000000 0.0000000
2 CM ~~ CM 0.0000000 0.0000000
7 MDD ~~ MDD 0.0000000 0.0000000
STD_All p_value
5 0.6697152 1.333828e-40
4 1.0000000 3.743973e-122
9 0.7426180 4.522471e-50
8 1.0000000 NA
1 1.0000000 NA
3 0.0000000 NA
6 0.0000000 NA
2 0.0000000 NA
7 0.0000000 NA

 my userGWAS model:

model<-'cm=~NA*MDD +start(0.2)*MDD + start(0.5)*CM

         ncm=~NA*MDD +start(0.2)*MDD

         

         cm~SNP

         ncm~SNP

         ncm~~1*ncm

         cm~~1*cm

         cm~~0*ncm

         CM ~~ 0*MDD

         CM~~0*CM

         MDD~~0*MDD

         SNP~~SNP'

my LDSC output:
Estimating heritability [1/3] for: CM.sumstats.gz
Heritability Results for trait: CM.sumstats.gz
Mean Chi^2 across remaining SNPs: 1.1537
Lambda GC: 1.1375
Intercept: 1.0115 (0.0075)
Ratio: 0.0746 (0.0489)
Total Observed Scale h2: 0.0566 (0.0048)
h2 Z: 11.751823
Calculating genetic covariance [2/3] for traits: CM.sumstats.gz and MDD.sumstats.gz
1058952 SNPs remain after merging CM.sumstats.gz and MDD.sumstats.gz summary statistics
Results for genetic covariance between: CM.sumstats.gz and MDD.sumstats.gz
Mean Z*Z: 0.1481
Cross trait Intercept: 0.0158 (0.0062)
Total Observed Scale Genetic Covariance (g_cov): 0.0259 (0.0021)
g_cov Z: 12.553063
g_cov P-value: 3.8242e-36
Estimating heritability [3/3] for: MDD.sumstats.gz
Heritability Results for trait: MDD.sumstats.gz
Mean Chi^2 across remaining SNPs: 1.2522
Lambda GC: 1.2242
Intercept: 0.9935 (0.0081)
Ratio: -0.0258 (0.0322)
Total Observed Scale h2: 0.0265 (0.0015)
h2 Z: 17.105943