Title page for ETD etd-05112007-200728

Type of Document

Dissertation

Author

Mei, Hao ,

Author's Email Address

hmei@ncsu.edu

URN

etd-05112007-200728

Title

Novel Methods for Mapping Complex Disease

Degree

PhD

Graduate Program

Bioinformatics

Advisory Committee

Advisor Name	Title
Zhao-Bang Zeng	Committee Chair
Eden R. Martin	Committee Co-Chair
Greg Gibson	Committee Member
Jung-Ying Tzeng	Committee Member
Margaret A. Pericak-Vance	Committee Member

Keywords

Dimension Reduction
Gene-Gene Interaction
Genetic Heterogeneity
Gene Mapping

Date of Defense

2007-05-11

Availability

unrestricted

Abstract

In contrast to simple disease (or Mendalian disease), complex disease has its special challenging characteristics for mapping. Firstly, complex disease generally does not have clear pattern of inheritance. Secondly, high-order interaction often occurs in complex disease, where variants are widely assumed to be common (Common Disease/Common Variants assumption) with only small or modest effects of each variants. Thirdly, genetic effects are often heterogeneneous, where different affected individuals may attribute to different sets of causative genes. These characteristics cause low power of mapping complex disease gene using linkage and tradtional association methods,which initiate us to develop powerful noval methods with high computational efficiency.

The first method extended the traditional method of Multifactor Dimensionality Reduction (EMDR) to detect high-order interaction by finding significant multi-locus models. EMDR does not assume any pattern inheritance. It applies the technique of dimension reduction and goodness-of-fitness test in the whole data to measure association from a multi-locus model by chi^2 statistic. This procedure is called non-crossvalidation in contrast to 10-fold crossvalidaiton by MDR. The significance of chi^2 statistic is tested by non-fixed permutation in contrast to omnibus permutation by MDR. By testing data from Genetic Analysis Workshop 14 (GAW14) with known answers, it was shown that EMDR with non-crossvalidation and non-fixed permutation is more powerful than MDR without increasing type I error. In addition, computationally high efficiency of developed EMDR program makes analyzing large number of markers simultaneous possible.

However, EMDR does not consider genetic heterogeneity,which could decrease an association signal in the data. The second method, MDR-Phenomics, is developed by integrating phenotypic information in the analysis of EMDR. Since genetic heterogeneity is often reflected by phenotypic heterogeneity, it makes possible that genetic heterogeneity can be controlled by analyzing phenotypic covariate. MDR-Phenomics classifies data into different groups based on discrete levels of a phenotypic covariate. The different association across classified groups is measured by F statistic using ANOVA method. A new M statistic measuring association from multi-locus model corrects possible decreased association signal due to genetic heterogeneity by multiplying EMDR chi^2 with F statistic.The significance of the M statistic is tested by permutation method. In tests of simulated data sets shows that MDR-Phenomics is powerful under genetic heterogeneity, and analysis of MDR-Phenomics in autism data successfully detected significant 2-locus model indicating potential interaction between serotonin transporter gene [SLC6A4] and integrin beta 3 [ITGB3] on chromosome 17.

Genetic heterogeneity and interaction indicate that a subset may exist with a homogeneous genetic effect, where an allele of a locus close to a causative variant is over tansmitted from parent to affected individuals (i.e., positive transmission). Based on that, Phenotypic Homogeneity Distinction (PHD) method is developed to find a phenotypic IDENTIFIER (ID), by which a subset can be identified with decreased genetic heterogeneity for association study. Such conditional association study is expected to increase power of mapping complex gene compared to traditional association methods. PHD takes two steps, an existence test and a definition procedure, to obtain a phenotypic ID by analysis of phenotypic covariate. Different strategies and statistic methods are proposed in the two steps for both categorical and continuous phenotypic covariate. To evaluate those strategies and methods, many data sets were simulated for tests, and results demonstrated that conditional association study on ID is more powerful than traditional association method.

Files

Filename		Size	Approximate Download Time (Hours:Minutes:Seconds)
Filename		Size	28.8 Modem	56K Modem	ISDN (64 Kb)	ISDN (128 Kb)	Higher-speed Access
	etd.pdf	527.77 Kb	00:02:26	00:01:15	00:01:05	00:00:32	00:00:02