Comparison of statistical methods of haplotype reconstruction and logistic regression for association studies

Investigating association between disease and single nucleotide polymorphisms (SNPs) has been an approach for genetic association studies and more recently investigating association between disease and haplotypes has become another accepted method. Haplotypes are physically linked combinations of alleles from a stretch of DNA and can serve to increase power of finding an association due to interactions between inclusive SNPs and the increased area of chromosome that is taken into consideration. Determining haplotypes experimentally or by family studies is a costly and timeinefficient method, so haplotype reconstruction by statistical methods has become an adopted practice. The problem with computational methods is the extra. source of error from ambiguous haplotypes that has to be included in statistical analysis. This paper investigates methods of error management with three different 1ogistic regression packages, two of which are specific to analysis of genetic data. Methods are applied to simulated data and a data set looking for genetic risk factors for non-Hodgkin Lymphoma.