Skip to content

Genomic SEM

Summary

Genomic SEM is an R package that fits structural equation models to the joint genetic architecture of multiple traits, using only GWAS summary statistics (no individual-level data required). It estimates shared genetic factors across correlated phenotypes, tests SNP effects on those latent factors, and builds polygenic scores from the resulting multivariate model — used, among other applications, to identify a shared genetic liability factor across psychiatric disorders.

Method

Genomic SEM operates in two stages.[1] First, it estimates an empirical genetic covariance matrix and its associated sampling covariance matrix via a multivariable extension of LD-score regression: diagonal entries are SNP heritabilities, off-diagonal entries are genetic covariances between trait pairs, and the sampling covariance matrix explicitly accounts for unknown/partial sample overlap across the contributing GWAS. Second, a user-specified structural equation model (e.g. a common-factor model) is fit to this genetic covariance matrix by minimizing the discrepancy between the model-implied and empirical covariance structures.[1] This lets Genomic SEM model multivariate genetic associations among phenotypes, identify variants acting on a general dimension of cross-trait liability, and construct polygenic scores from the fitted latent factor(s), without needing raw genotype/phenotype data from every contributing study.

Application: Joint Analysis of Five Psychiatric Traits

Fitting a single common "p-factor" to European-ancestry summary statistics for schizophrenia, bipolar disorder, major depressive disorder, PTSD, and anxiety, the authors identified 128 independent genome-wide-significant loci for the p-factor, of which 27 were not significant in any single contributing univariate GWAS (5 of these 27 were subsequently validated in separate published GWAS at P < 1×10⁻⁵). Schizophrenia and bipolar disorder showed the strongest loadings onto the genetic p-factor (model fit: χ²(5) = 89.55, CFI = 0.848, SRMR = 0.212).[1] Polygenic scores built from the p-factor out-predicted every univariate polygenic score for psychiatric symptoms in UK Biobank (depression, psychotic experiences, mania, anxiety, PTSD symptoms, and a phenotypic p-factor) and for neuroticism in Generation Scotland.[1]

See Also

Citations

[1] Grotzinger, A. D., Rhemtulla, M., de Vlaming, R., Ritchie, S. J., Mallard, T. T., Hill, W. D., Ip, H. F., Marioni, R. E., McIntosh, A. M., Deary, I. J., Koellinger, P. D., Harden, K. P., Nivard, M. G., & Tucker-Drob, E. M. (2019). Genomic structural equation modelling provides insights into the multivariate genetic architecture of complex traits. Nature Human Behaviour, 3, 513–525. DOI: 10.1038/s41562-019-0566-x. Source: grotzinger2019-genomic-sem.md. Supports: method description and psychiatric-trait application above. Location: Full text — Methods and Results.