A two-stage classification approach identifies seven susceptibility genes for a simulated complex disease.

The simulated data set of the Genetic Analysis Workshop 15 provided affection status, four quantitative traits, and a covariate. After studying the relationship between these variables, linkage analysis was undertaken. Analyses were performed in the first replicate only and without any prior knowledge of the underlying model. In addition to the main effect of the DR locus on chromosome 6, significant linkage was also identified on chromosomes 8, 9, 11, and 18. Notably, the power to detect linkage increased after transforming the skewed and kurtotic IgM and anti-CCP distributions. Moreover, genes on chromosome 11 could not be discerned from noise without the transformation, thus highlighting the need in real life situations for careful examination of the phenotypic data prior to genetic analysis. Significant association with one single-nucleotide polymorphism was identified for the regions on chromosome 11 and 18. Haplotype analyses were attempted for the other regions, but only the underlying variation of the DR locus could be identified. Two methods were then applied to predict classification using the factors identified so far. These methods - logistic regression and multifactor dimensionality reduction (MDR) - performed comparably for this data set. Those affected individuals that were misclassified as unaffected were then used in a genome-wide association analysis to identify additional susceptibility loci. Two additional loci were identified in this fashion, illustrating the usefulness of this two-stage classification approach.