by Tsai, Hui-Ju

Abstract (Summary)
Complex diseases such as type 2 diabetes, hypertension and psychiatric disorders have been major public health problems in US. In order to increase the power in the linkage analysis of complex traits, genetic heterogeneity has to be taken into account. During the past few years, several methods have been proposed for dealing with this issue by incorporating covariate information into the affected sib pair (ASP) analysis. However, it is still not clear how these approaches perform under different gene-environment (G x E) interactions. The covariate statistics evaluated in this study are: (1) mixture model; (2) general conditional-logistic model (LODPAL); (3) multinomial logistic regression models (MLRM under no dominance, no additive and min-max restriction); (4) extension of the maximum-likelihood-binomial approach (MLB); (5) ordered-subset analysis (OSA with three different rank orders: high-to-low, low-to-high and optimal-slice); (6) logistic regression modeling (COVLINK). Based on the chromosome-based approach, we have written simulation programs to generate data under various G x E models and disease models. We first define the empirical statistical significance thresholds using C2, the environmental risk factor, under the null hypothesis. We then evaluate the power of the covariate statistics when different covariates are used. We also compare the performance of the covariate statistics with the model-free methods (Sall and Spair). In all three G x E interaction models, most covariate methods perform better when using C1, the covariate with G x E interaction effect, than when using C2 or the random noise covariate C3, except for MLB and the low-to-high OSA method. Comparing with the model-free methods (using Sall as the baseline), mixture model and the high-to-low OSA method perform the best of the covariate statistics when using C1. However, when using C2 or C3, most covariate statistics provide less power than Sall. Only MLB has comparable power to Sall across all genetic models. According to our results, in different G x E interactions, one should apply the appropriate covariate statistic and include the suitable type of covariates carefully.
Bibliographical Information:

Advisor:Robert E. Ferrell; Eleanor Feingold; Bernie Devlin; M. Michael Barmada; Daniel E. Weeks

School:University of Pittsburgh

School Location:USA - Pennsylvania

Source Type:Master's Thesis

Keywords:human genetics


Date of Publication:04/29/2004

© 2009 All Rights Reserved.