A Statistical Framework for Expression Quantitative Trait Loci Mapping

Abstract

In 2001, Sen and Churchill reported a general Bayesian framework for quantitative trait loci (QTL) mapping in inbred line crosses. The framework is a powerful one, as many QTL mapping methods can be represented as special cases and many important considerations are accommodated. These considerations include accounting for covariates, nonstandard crosses, missing genotypes, genotyping errors, multiple interacting QTL, and nonnormal as well as multivariate phenotypes. The dimension of a multivariate phenotype easily handled within the framework is bounded by the number of subjects, as a full-rank covariance matrix describing correlations across the phenotypes is required. We address this limitation and extend the Sen-Churchill framework to accommodate expression quantitative trait loci (eQTL) mapping studies, where high-dimensional gene-expression phenotypes are obtained via microarrays. Doing so allows for the precise comparison of existing eQTL mapping approaches and facilitates the development of an eQTL interval-mapping approach that shares information across transcripts and improves localization of eQTL. Evaluations are based on simulation studies and a study of diabetes in mice.T HE quantitative trait loci (QTL) mapping framework developed by Sen and Churchill (2001), referred to hereinafter as the Sen-Churchill framework, unifies many methods for QTL mapping in inbred line crosses. The seminal work of Lander and Botstein (1989) and subsequent methods including Haley-Knott regression (1992), composite interval mapping, and multiple QTL mapping ( Jansen 1993;Zeng 1993Zeng , 1994Jansen and Stam 1994), are all represented, at least approximately, as special cases of the framework. The framework also accounts for covariates, nonstandard cross designs, missing genotype data, genotyping errors, multiple interacting QTL, and nonnormal as well as multivariate phenotypes. As a result, it provides a powerful approach to localize the genetic basis of quantitative traits.There has been much interest recently in identifying the genetic basis of thousands of gene-expression traits measured via microarrays (Brem et al. 2002;Schadt et al. 2003;Yvert et al. 2003;Cox 2004). The multi-trait version of the Sen-Churchill framework is based on the multivariate normal distribution. This approach becomes problematic when the number of traits is larger than the number of subjects, as the estimated covariance matrix will have less than full rank. To address this, we here extend the Sen-Churchill framework to accommodate expression phenotypes. We first highlight aspects of the Sen-Churchill framework important to our development, and then detail the extension. We show that the extended framework generalizes the currently available expression QTL (eQTL) mapping methods and facilitates the development of an approach that allows for both interval mapping of eQTL and information sharing across transcripts. Evaluations are based on simulation studies and a study of diabetes in mice. Generalizations of the framework are also discuss...

Expression Quantitative Trait Loci Mapping With Multivariate Sparse Partial Least Squares Regression

“…In Chun and Keles x (2007), this property is illustrated with as many as 5000 variables. Another important research question in eQTL mapping is allowing for linkages with locations between markers using interval mapping (Chen and Kendziorski 2007). Our current formulation allows for mapping only at exact marker locations.…”

Flor Revisited (Again): eQTL and Mutational Analysis of NB-LRR Mediated Immunity to Powdery Mildew in Barley

“…Transcript abundance of a single gene is a quantitative trait and its regulation can be genetically interrogated. This is often called genetical genomics, or eQTL mapping because the phenotypes in question are the expression of individual genes (Jansen and Nap 2001;Rockman and Kruglyak 2006;Chen and Kendziorski 2007). Thus, while comparative analysis of expression data has led to elucidation of significant co-expression networks, identification of the regulators of those networks has been limited; here, the use of eQTL analysis is advantageous.…”

Systems genetics, bioinformatics and eQTL mapping

“…Current methods for eQTL mapping rely mostly on either a microarray analysis procedure or a QTL mapping statistic since no well statistical method emerged. The Sen-Churchill method based on Bayesian framework for quantitative trait loci (QTL) was extended to accommodate eQTL mapping studies, where high-dimensional gene-expression phenotypes were obtained via microarrays (Chen and Kendziorski 2007). As pointed out in a number of articles on microarray data analysis (Newton et al 2001;Tusher et al 2001;Kendziorski et al 2003), information sharing is important to improve sensitivity and moderate test statistics that are otherwise prone to inflated error.…”