LPA to SNP association results across multiple disorders may be a promising approach to identify clusters of SNPs that are associated with related phenotypes. While the cross-tabulations of class memberships between 2- to 4-class models matched our expectation, some instability of class memberships were observed between LPA models with different number of classes. For example, 5.7–6.6% of SNPs in the Null class were classified into non-null classes, and 6.3–10.5% of SNPs in non-null classes were classified into the null class with increasing number of classes. Because Table 3 is based on the most likely memberships, SNPs with unstable memberships are likely to be in the edge of the given class where it overlap with other classes. Decreasing entropy values with increasing number of classes in Table 1 may indicate that the overlap between classes increases as the number of classes increase.
