Finding Quantitative Trait Genes in the Mouse

Lead Research Organisation: University of Oxford
Department Name: Wellcome Trust Centre for Human Genetics


Genes interact with one another, and statistical methods exist to test for them, but current genetic studies to detect genes which contribute to disease risk do not consider the possibility of gene interactions. This is largely because there are so many different possible interactions to be considered, and a huge number of comparisons are needed to test for all of them. However if interactions and their effects are ignored, some genes may not be found because their effects are masked by effects of other genes, or because they would have no effect on their own. I will address the need for new analysis methods to cope with these problems and develop software for general use. I will apply statistical methods which consider all possible gene interactions, without testing each separately, to identify new disease genes and their interactions. This approach can exploit the growing body of biological information on how genes behave to narrow down the number of likely possibilities to consider in the analysis. This approach will be an important addition to existing methods of genetic analysis, providing a systematic and user-friendly method of searching for gene interactions and discovering new genes affecting human disease.

Technical Summary

The aim of this project is to develop methods to identify quantitative trait genes by modelling interactions (epistasis) between loci contributing to quantitative trait phenotypes. Analyses which ignore epistasis may miss relevant genes, particularly if epistasis substantially alters the expression of the genes involved. However, there are no reliable estimates of the extent of epistasis, and because the network of gene effects and interactions is almost always unknown, and may be extremely complex, there are a great many possibilities to consider when modelling epistasis.
I propose an integrated approach, where statistical interactions (based on genetic mapping data) and functional interactions (based on annotations) are combined in a joint model. Within a Bayesian statistical framework, I can allow for uncertainty about the true number loci and interactions involved. The unknown network of genes can be inferred statistically from genomic data using a model-averaging approach. Bayesian model averaging has been successfully employed to find QTL in simple mouse crosses. I will analyse a dataset of outbred heterogeneous stock mice, of sufficient size to detect QTL with small effect, and suitable for high resolution mapping. This type of population has some similarities with, and shares many of the issues found in data from human populations, leading to an analytical approach which could be extended to human data.
This method allows incorporation of prior knowledge to aid the statistical model inference, including information on network structure and motifs from systems biology and mathematical graph theory, and functional data from sequence annotation databases. This will aid prioritisation of the list of candidate genes identified in a genome-wide analysis and narrow the range of models needing to be considered, resulting in an increase in power for these studies. I will also develop user-friendly software which will be made available for general use, to discover new quantitative trait genes for complex phenotypes.


10 25 50
Title HBREM 
Description I have developed a novel hierarchical Bayesian random effect model (HBREM) for QTL mapping, which does not require Markov chain Monte Carlo techniques, applicable to known or inferred haplotype data, either from human genotype data or, in model organisms, from a panel of recombinant inbred lines constructed from multiple founder inbred lines where the mosaic of founder lines has been reconstructed. 
Type Of Material Data analysis technique 
Year Produced 2009 
Provided To Others? Yes  
Impact The method provides more accurate estimates of individual haplotype and haplotype pair effects without any loss of overall power. Existing methods do not provide reliable estimates of individual effects for inferred haplotype data with many possible haplotypes. Individual effect estimates are important when investigating a significant QTL to discover which haplotypes are responsible for the signal and hence which variant(s) may be responsible and the mechanism. 
Description Arabidopsis thaliana MAGIC lines 
Organisation University of Manchester
Department Faculty of Life Sciences
Country United Kingdom 
Sector Academic/University 
PI Contribution The MAGIC lines are a panel of 19 recombinant inbred lines developed to map traits in Arabidopsis thaliana. I am a member of the analysis team in my supervisor's group and have applied the Bayesian methods I developed during my MRC fellowship.
Collaborator Contribution Provision of data for analysis Authorship on paper Kover et al 2009 PLoS Genetics
Impact I presented posters at the annual CTC meetings in Montreal (2008) and Manchester (2009) and contributed to a publication of the analysed data (PubMed ID 19593375).
Start Year 2008
Description Depression GxE interaction 
Organisation University of Bristol
Department School of Experimental Psychology
Country United Kingdom 
Sector Academic/University 
PI Contribution I conducted some simulation and statistical analysis on a particular gene X environment interaction for which a meta-analysis was carried out.
Impact Two publications on a gene X environment interaction for depression (one in press, PubMed ID 18691701 for the other)
Start Year 2007
Description Mouse Collaborative Cross (CC) 
Organisation University of North Carolina at Chapel Hill
Country United States 
Sector Academic/University 
PI Contribution The CC is an international project to generate a genetic reference panel of over 500 recombinant inbred lines of mice. I am a member of the analysis team in my supervisor's group, and have applied the Bayesian methods I developed during my MRC fellowship to map QTLs in this population.
Collaborator Contribution Provision of data for analysis
Impact I presented posters at the annual CTC meetings in Montreal (2008) and Manchester (2009)
Start Year 2008