Academic journal article Genetics

Bayesian Multiple Quantitative Trait Loci Mapping for Complex Traits Using Markers of the Entire Genome

Academic journal article Genetics

Bayesian Multiple Quantitative Trait Loci Mapping for Complex Traits Using Markers of the Entire Genome

Article excerpt

ABSTRACT

A Bayesian methodology has been developed for multiple quantitative trait loci (QTL) mapping of complex binary traits that follow liability threshold models. Unlike most QTL mapping methods where only one or a few markers are used at a time, the proposed method utilizes all markers across the genome simultaneously. The outperformance of our Bayesian method over the traditional single-marker analysis and interval mapping has been illustrated via simulations and real data analysis to identify candidate loci associated with colorectal cancer.

(ProQuest: ... denotes formulae omitted.)

TREMENDOUS advances have been achieved over the last decade in the identification of genes underlying many heritable traits with the greatest progress limited almost entirely to those with Mendelian inheritance patterns and well-defined quantitative traits that have relatively large and consistent effects. However, many common pathologies afflicting the greatest number of individuals are not due to simple Mendelian traits. Recent emphasis has been shifted to map complex traits, which are caused by the sum of a complex interaction between gene products and environmental stimuli. Complicating the analysis of these types of traits is the prediction that many are also controlled by genes that have small effects individually, but whose cumulative action is the cause of significant interindividual variation. Due to the complex and often subtle nature of phenotypic variation, traits with complex etiologies have proven far more resistant to genetic analysis. Most of the available quantitative trait loci (QTL) mapping methods map only one or a few QTL at a time and therefore are not efficient for mapping such complex traits. Forward and stepwise selection procedures have been proposed in searching for multiple QTL. Though simple, these methods have their limitations, such as the uncertainty of number of QTL, the sequential model building that makes it unclear how to assess the significance of the associated tests, etc.

To overcome this problem, Bayesian QTL mapping (SATAGOPAN et al. 1996; SILLANPAA and ARJAS 1998; STEPHENS and FISCH 1998; YI and XU 2000, 2001; HOESCHELE 2001) has been developed, in particular, for detection ofmultipleQTL by treating the number of QTL as a random variable and specifically modeling it using reversible-jump Markov chain Monte Carlo (MCMC) (GREEN 1995). Due to the change of dimensionality, care must be taken in determining the acceptance probability for such a dimension change, which in practice may not be handled correctly (VEN 2004). To avoid such a problem by the uncertain dimensionality of parameter space, YI (2004) and XU (2003) proposed a unified Bayesian framework to identify multiple QTL using all markers across the genome. The method of XU (2003) is based on a shrinkage idea to simultaneously evaluate marker effects of the entire genome under the random regression model by assigning each marker a normal prior with mean 0 and an effect-specific variance. The effect-specific prior variance was further assigned a vague prior such that the variance was estimated from the data. Those markers that have no effect on the trait will be essentially shrunk down to 0. Similarly, YI (2004) adapted the stochastic search variable selection (SSVS) approach of GEORGE and MCCULLOCH (1993) to the QTL mapping framework. SSVS is a variable selection method that keeps all possible variables in the model and limits the posterior distribution of nonsignificant variables in a small neighborhood of 0 and therefore eliminates the need to remove nonsignificant variables from themodel. In principle, XU (2003) and YI (2004)'s methods are similar and both have the ability to control the genetic variances of a large number of QTL where each has small effect (WANG et al. 2005). Due to the simplicity of XU (2003), we decide to go with the unified shrinkage method in this article.

Some quantitative traits do not have continuous measurements, but rather are qualitative traits with, for example, binary measurements. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.