Improvement of the power to detect complex disease genes by regional inference procedures.

Theoretical studies and simulations suggest that "true" linkage peaks are longer than "false" peaks of the same significance level. Our goal for this study was to improve the power of linkage detection by using a regional criterion for linkage; that is, requiring more than one p-value in a given region to pass a threshold. We tested this method by determining the power and type I error for finding the underlying loci on chromosomes 5 and 8 that contribute to the variability of Q1 (after adjusting Q1 for covariates). We used the Haseman-Elston sib-pair statistic to test for linkage of all 367 markers to the adjusted Q1 trait in 100 replicates. We compared the regional inference procedure to that of the Lander and Kruglyak (LK) criteria for significant and suggestive linkage. For example, the power to detect the chromosome 5 locus was 48% for the LK criterion for significant linkage (p < or = 0.0001) and 63% when we required two p-values out of five consecutive ones to be < or = 0.001. The type I error was not more than 5% for either method (2% for the LK and 5% for our criterion). This suggests that using a criterion based on length may improve the power of linkage detection for complex traits.