Statistical Methods for Linkage and Association Analyses Albert Kingman Z01DE00689 As genome scans used for genetic mapping studies for complex diseases expand to the use of thousands of markers, too many regions are identified in which modest p-values are realized but fail to qualify for either suggestive or significant linkage by the linkage criteria of Lander & Kruglyak, 1995. One possible approach is to combine information of linkage from clusters of correlated markers into 'regional' linkage tests. Two such classes of 'regional' linkage test statistics were investigated for their potential to increase power in linkage tests for a QTL over a wide range of additive heritability, and different marker locations. The 1st class incorporated min t statistics, which were minimums of consecutive H-E t-test statistics; the 2nd set incorporated moving averages, ma(t), of the H-E t-test statistics. The Genometric Analysis Simulation Program (GASP) was used to simulate the samples. Two distinct chromosomal segments were simulated in these experiments under several genetic models. Each chromosomal segment contained 20 markers uniformly distributed at 10-cM intervals. A single locus, responsible in part for the phenotypic variation of a quantitative trait, was placed at one of three different locations on the 1st chromosomal segment: at the left edge (between marker loci 1 and 2), at the 1st quartile (between marker loci 5 and 6) and near the center (between marker loci 10 and 11). Separate simulations were run for each trait location. The results showed that real gains in power, ranging between 5% and 20% were realized by the moving average marker-cluster statistics compared with the single locus t-test. The greatest gains were for additive heritability models ranging from 30% to 60%. The gain in power of the moving average of 2 consecutive t-tests over a single t-test was comparable achieving an equivalent power for the t-test in detecting a trait having a 10% smaller heritability level for the same study design. The min t statistic based on pairs of consecutive t-test statistics also produced similar gains in power, but the min t statistics based on 3 or more consecutive t-test statistics produced losses in power. Thus, the potential for real gains in power in dense genome screens may be realized by the implementation of some marker-cluster based test statistics. In a follow-up study we investigated the stochastic properties of p-values associated with the H-E Sib-pair linkage test under the alternative hypothesis of linkage. The purpose of this investigation is to relate the expected p-values (EPV) to the significance level and power for this test procedure, demonstrating irts functional relationship with trait heritability and distance of marker from the trait. Studies are also underway investigating the heritability and anticipation levels of Hodgkin's disease and non-Hodgkin's lymphoma, using a Swedish Cancer Registry. Another study investigating whether or not the occurrences of a specific sequence of 8-nucleotides appears in clusters as compared with a random pattern are in progress. Specific statistical tests of randomness are being used and compared.
