full set of SNPs. To incorporate common variance within classes, we fit the 3-class model with within-class residual covariance between indicators allowed. The result indicated no noticeable change in class proportions and profiles. Estimated residual correlations were substantial especially between AD and ASP in the Null and MD classes, 0.205 and 0.223, respectively, and between MD and ASP, in the MD and AD+ASP classes, 0.177 and 0.181, respectively. Although our analysis indicated an unexplained source of correlations between indicators, the stability of results with or without residual covariance may indicate that the classes we identified are stable regardless of local independence assumption.
