Fig. 4: Genomic SEM one-factor and two-factor model.
From: Phenotypic and genetic markers of psychopathology in a population-based sample of older adults
Confirmatory factor analyses were conducted on the GWAS summary statistics in Table 1, using the Genomic SEM package in R Statistical Software (Grotzinger et al., 2019). Standardized estimates are shown. See Supplemental Fig. 3 for unstandardized estimates. In both the one-factor and two-factor models, the residual variance of MDD was fixed to zero. Model fit comparisons between the one-factor model (χ²(14) = 76.762, p < 0.001, AIC = 104.762, CFI = 0.962, SRMR = 0.127) and two-factor model (χ²(13) = 46.072, p < 0.001, AIC = 76.072, CFI = 0.980, SRMR = 0.084) indicated superior model fit of the two-factor model (Δχ² = 30.69(1), p < 0.001, ΔCFI > 0.01, lower AIC). Single-nucleotide polymorphism effects were then integrated into the model to derive new SNP weights for the construction of latent polygenic scores (see Supplemental Methods).
