During this fiscal year, this project, originally begun when Dr. Bailey-Wilson was a Professor at Louisiana State University Medical Center, has been continued. The project is designed to use computer simulation techniques to determine the power and robustness of several types of linkage analysis to various failures of assumptions.
The aims of the study are to 1) evaluate the Type I error rates of non-parametric Haseman-Elston sib-pair linkage analysis of both quantitative and qualitative traits (with and without environmental covariates) to misspecification of marker locus allele frequencies when parental data are missing; 2) to determine and compare the power of both Haseman-Elston sib-pair linkage and parametric lod-score linkage for both quantitative and qualitative traits under a variety of misspecifications of the trait and marker loci; 3) to eventually include other methods of linkage analysis in these comparisons. To date, Type I error rates for a quantitative trait with high heritability and no environmental covariates have been determined for the H-E test and the parametric Lod- score linkage test when marker alleles are misspecified. In general the H-E test is very robust to this sort of misspecification when using modern marker loci (at least 5 alleles with high heterozygosity) but lod-score linkage tests are not robust when parental marker genotypes are missing. Power to detect a linked marker using this same type of quantitative trait has been determined for both the H-E and the lod-score tests when marker allele frequencies are misspecified. When modern markers are used, the effect of misspecification on power is small for both methods, with the effect often being less pronounced for the H-E test. We have also shown that power of the lod-score method is drastically reduced by only small to moderate misspecifications of trait locus genotypic means. We have also shown that very small misspecifications in the trait model combined with small to moderate misspecifications in marker allele frequencies are sufficient to cause large increases in Type I error. We have further shown that the H-E test is robust to misspecification of marker allele frequencies when parental data are missing for a wider variety of quantitative traits. Our initial studies showed this was true for a quantitative trait with high heritability. Our further work has shown that this test is remarkably robust for traits with only moderate heritability and also for traits that are completely environmentally determined. This is important because there has been concern that there might be particular inflation of Type I error rates for traits without a genetic component. This study showed that this is not a concern using modern highly polymorphic markers and the H-E test. We have also further examined the effects of allele frequency misspecification on model-based lod-score linkage. We have found that when the trait model is correctly specified, parental marker genotypes are unknown, and marker allele frequencies are misspecified, Type I error rates of lod score linkage are inflated for traits with large to moderate heritability, but this test is robust when the trait is completely environmentally determined. One paper based on these results was published in this fiscal year. We have further determined that the cause of the inflation of Type I error rates is partially due to sampling variation. Some samples drawn from a population are outliers and do not reflect the underlying population accurately. In these types of samples, using the population parameters in the linkage analysis leads to a model that does not fit the data well, thus leading to increases in the number of false positive linkage results. In this fiscal year, we have performed extensive simulations exploring ways to resolve this problem, including performing segregation analysis on each simulated dataset prior to linkage studies. This approach does reduce the inflation of the false positive rate. A manuscript describing these results is in preparation. Currently, we are pursuing two projects designed to examine the effects of important environmental covariates on power and Type I error in linkage studies. We are simulating traits for which moderate to strong environmentsl risk factors play a role in risk of a disease trait (affected vs. unaffected) and then comparing the performance of various analytic methods that ignore covariates to the performance of methods that incorporate these covariates into the analysis.
| Mandal, D M; Wilson, A F; Elston, R C et al. (2000) Effects of misspecification of allele frequencies on the type I error rate of model-free linkage analysis. Hum Hered 50:126-32 |
| Mandal, D M; Sorant, A J; Pugh, E W et al. (1999) Environmental covariates: effects on the power of sib-pair linkage methods. Genet Epidemiol 17 Suppl 1:S643-8 |
