| For complex diseases, we often sample and genotype affected sibships to map the disease of interest by linkage analysis. Subsets of these data then may be used for association studies. Often these association studies use only one affected individual per family in a case-control design, even when all family members are genotyped. Moreover, the selection of one individual per family for a case-control study is arbitrary, with different choices resulting in different samples and variation in results. I introduce a method to test for disease-marker association that uses the information from all affected sibs simultaneously.; In haplotype-based association studies for late onset diseases, one attractive design is to use unaffected spouses as controls. Given cases and spouses only, the EM algorithm for case-control data can be used to estimate haplotype frequencies. But often we have offspring for some spouse pairs and their genotypes provide additional information. Existing methods either ignore the offspring information, or reconstruct parental haplotypes using offspring information and discard data from those whose haplotypes cannot be reconstructed with high confidence. Neither approach is efficient, and the latter may also be biased. For such data, I propose a unified, likelihood-based method of haplotype inference, which uses available offspring information to apportion ambiguous haplotypes for the parents. I also describe some tests for disease-haplotype association.; Given a disease-associated marker allele, we ask if this allele, or one in linkage disequilibrium with it, can account partly for the linkage signal at the marker. We address this question by determining if families selected based on presence of the allele show stronger evidence of linkage. For affected sibship data, I describe three subsetting schemes, and introduce procedures to test if the linkage signal in a region can be partly attributed to the presence of an associated allele. The tests are powerful even for disease models with sib relative risk λS = 1.1, and even for data with no evidence for linkage due to sampling variation.; These three new approaches extend existing methods to provide more efficient use of collected data as well as more powerful test for detecting disease genes through association analyses. |
