We applied LPA to the log of p-values from the GWAS of AD, ASP, and MD symptom counts. LPA is a categorical latent variable model that is commonly used to identify groups of individuals, referred to as “classes”, based on the patterns of responses on multiple observed variables, referred to as “class indicators”. In this study, however, we used LPA to identify classes of SNPs, instead of individuals. Thus, unlike the common setting of LPA, in which data are arranged by individuals, the data for our analyses were arranged by SNPs. Thus, the data consisted of four columns—a SNP ID and three columns of class indicators, which in our case are −log10(p-values) corresponding to AD, ASP, and MD, with n rows, where n is the number of SNPs included in the analysis. The profile of each SNP, the pattern of association strengths with the three phenotypes, indexed by their −log10 of GWAS p-values, was used as class indicators in LPA.
