Figure 1

Parameter space for association study. The first dimension (in black) applies to both a general linear model and mixed linear model (MLM). The other dimensions apply to MLM only. The population structure (Structure) is fitted as a fixed effect with effect estimated as the best linear unbiased estimates (BLUE). The second dimension introduces individuals as random effects with variance defined by a kinship matrix. The best linear unbiased prediction (BLUP) for random effects can be solved directly with known variance components. The third dimension estimates unknown variance components using algorithms such as the residual maximum likelihood algorithm. The fourth dimension clusters individuals into groups (compression) by using cluster algorithms. The fifth dimension determines the best number of groups or average number of individuals per group (defined as compression level). The current study developed a sixth dimension that determines the best algorithm to define group kinship, for example, average, median, or maximum. The two dimensions in red belong to the standard MLM based on the individuals. The remaining dimensions (in blue) belong to the compressed MLM based on groups.