Running usermodel

This video is a technical tutorial on using the ‘usermodel’ function within the GenomicSEM R package for multivariate genome-wide association studies (GWAS). The presenter begins by reviewing the motivation for GenomicSEM, which extends bivariate LD score regression to model complex genetic correlations across multiple traits. They then explain the lavaan syntax used to specify structural equation models, including regressions, covariances, factor loadings, and parameter constraints. The video demonstrates the two required arguments for usermodel: the LDSC output (containing the genetic covariance matrix S and sampling covariance matrix V) and the model specification. Key optional arguments such as std.lv (fixing latent variances to 1) and imp_cov (requesting implied and residual covariance matrices) are discussed. The presenter then walks through a practical example fitting a common factor model to four psychiatric traits (MDD, PTSD, alcohol, anxiety), showing how to interpret the output including factor loadings, residual variances, and model fit indices. They emphasize the importance of model fit indices like CFI and SRMR, cautioning about the sensitivity of chi-square to large GWAS sample sizes. The video concludes by mentioning extensions for gene expression and SNP-level analyses, setting the stage for subsequent tutorials.