Identifying gene-gene interactions associated with complex diseases and complex traits

Abstract

Diseases are usually associated with genetic variants, mainly single nucleotide polymorphisms (SNPs) or Single Sequence Repeat Polymorphisms (SSRPs). Therefore it's an important task for researchers in human genetics to search for genetic factors having influence on diseases as it can be used in many medical case-control studies. In recent years, this research has been greatly improved by using genome-wide association studies (GWASs) which use a single-locus approach, where each variant is tested individually for association with a specific disease. However most complex diseases are considered to be the results of gene-gene and gene-environment interactions. Many computational methods have been proposed to detect if a particular set of genes has epistatic interaction with a particular complex disease. However, even though many such methods have been shown to be useful, they can be made more effective if the properties of gene-gene interactions can be better understood. Towards this goal, we have attempted to uncover patterns in gene-gene interaction and the patterns reveal an interesting property that can be reflected in an inequality that describes the relationship between two genotype variables that takes on the genotypes of two different genes as values, and a disease-status variable that takes on binary values representing the presence or absence of a complex disease. We show that this inequality can be derived for generalization to n genotype variables. Based on this inequality, we establish a conditional independence and redundancy (CIR) based definition of gene-gene interaction and the concept of an interaction group. We discuss the properties of these concepts and explain how they can be used in a novel algorithm that can be used to detect gene-gene interaction with an order of two and above greater than two. Experimental results using both simulated and real datasets show that the proposed algorithm can be very promising. Possible ways to further improve the effectiveness of the new algorithm are also provided. Like complex diseases, complex quantitative traits (QTs) are also usually associated with genetic variants. The majority of innate and acquired body and behavioral characteristics. Many physiological characteristics are also reflected by complex traits. In addition, most diseases exhibit various symptoms through complex traits. The Multifactor Dimensionality Reduction (MDR) method was originally proposed as a nonparametric and model-free data reduction approach for identifying interactions without significant main effects and has been successfully applied to identify gene-gene interactions in many common complex diseases. Some efforts have been made to extend MDR to QTs. However these methods are still not computationally efficient or effective. Therefore we propose Extended Fuzzy Quantitative trait MDR (EFQMDR) to strengthen identification of gene-gene interactions associated with a quantitative trait by first transforming it to an ordinal trait and then using a balanced accuracy measure based on extended member functions of fuzzy sets to select multiple best sets of genetic markers as having strongest associations with the trait. Experimental results on simulated datasets and real datasets show that our algorithm has better performance in terms of test accuracy and consistency in identifying gene-gene interactions associated with QTs. Multiple correlated phenotypes often appear in complex traits or complex diseases. These correlated phenotypes are useful in identifying gene-gene interactions associated with complex traits or complex disease more effectively. Some approaches have been proposed to use correlation among multiple phenotypes to identify gene-gene interactions that are common to multiple phenotypes. However these approaches either didn't find truly gene-gene interactions or the results are hard to explain, especially using all correlated phenotypes to identify gene-gene interactions make identified interactions unreliable. Multivariate Quantitative trait based Ordinal MDR (MQOMDR) algorithm is therefore proposed to effectively identify gene-gene interactions associated with multiple correlated phenotypes by selecting the best classifier according to not only the training accuracy of the phenotype under consideration but also other phenotypes with weights determined mainly by their pair correlation with the phenotype under consideration and also by repeated selection process to make use of truly useful correlated phenotypes. Experimental results on two real datasets show that our algorithm has better performance in identifying gene-gene interactions associated with multiple correlated phenotypes.