Statistical properties of 3 model based tests for linkage

• Main Author: Ulgen, Ayse
• Format: Dissertation
• Language: English
• Published: ProQuest Dissertations & Theses 01-01-2001
• Subjects: Statistics
• ISBN: 0493365508; 9780493365503

The lod score was originally developed for evaluating linkage for simple Mendelian disorders. Today, most diseases are complex and we don't know the underlying genetic model and therefore it is appropriate to use maximum lod scores. In this paper, we consider mapping of a disease whose mode of inheritance is not known using maximum lod score statistic with five genetic parameters to be estimated; disease allele frequency, phenocopy coefficient, penetrance, dominance and the recombination fraction. We seek to determine the statistical properties of the maximum lod score and determine the distribution of the null distribution by a large simulation study. We have improved an existing numerical method, Shrinking Orthogonal Latin Hypercube Design (SOLH) (Ye, 1997 and Chen, 1998) to maximize the lod score over models and find the setting of its parameters. We use two-point linkage analysis to compute the lod scores using LIPED and maximize the lod score using the SOLH. We developed a technique based on ANOVA to determine the initial ranges of the parameters required in order to obtain the maximum lod score. We used three sets of 1200 replicates with three generating models to study the null distribution of the maximum lod score. The results suggest that the null distribution of the maximum lod score based on SOLH is invariant under three very different genetic generating models but the distribution of the combined sample of 3600 replicates does not appear to fit to any well known probability distribution. The parameter estimates obtained under the alternative hypothesis appear to be unbiased. A comparative power study shows that lod scores obtained using the generating model have only slightly higher power than those obtained through maximization.