In our analyses, we utilized three primary sets of parameters to interpret the results of LPA. First, the estimated means of the −log10(p) of the three phenotypes were used to characterize and interpret the response profile of each class. For example, if SNPs in a given class have relatively larger means for AD and ASP than for MD, this class can be interpreted as being associated with the two externalizing phenotypes. Second, LPA estimates each SNP’s probability of belonging to each of the estimated classes. For each SNP, these probabilities sum to one, and the class with the highest probability was referred to as the “most likely class.” Third, LPA computes class proportions, which are, in our case, the proportions of SNPs classified into each class, based on the estimated parameters. In addition, the quality of classification was measured by the entropy. Entropy ranges from 0 to 1, where values approaching 1 indicate clear separation between classes (Celeux and Soromenho 1996). We used Mplus version 7.1 (Muthén and Muthén 1998–2012), with maximum likelihood estimator with robust standard errors (MLR), to estimate the parameters of LPA.
