ClusterMI: Detecting High-Order SNP Interactions Based on Clustering and Mutual Information

Abstract

Identifying single nucleotide polymorphism (SNP) interactions is considered as a popular and crucial way for explaining the missing heritability of complex diseases in genome-wide association studies (GWAS). Many approaches have been proposed to detect SNP interactions. However, existing approaches generally suffer from the high computational complexity resulting from the explosion of candidate high-order interactions. In this paper, we propose a two-stage approach (called ClusterMI) to detect high-order genome-wide SNP interactions based on significant pairwise SNP combinations. In the screening stage, to alleviate the huge computational burden, ClusterMI firstly applies a clustering algorithm combined with mutual information to divide SNPs into different clusters. Then, ClusterMI utilizes conditional mutual information to screen significant pairwise SNP combinations in each cluster. In this way, there is a higher probability of identifying significant two-locus combinations in each group, and the computational load for the follow-up search can be greatly reduced. In the search stage, two different search strategies (exhaustive search and improved ant colony optimization search) are provided to detect high-order SNP interactions based on the cardinality of significant two-locus combinations. Extensive simulation experiments show that ClusterMI has better performance than other related and competitive approaches. Experiments on two real case-control datasets from Wellcome Trust Case Control Consortium (WTCCC) also demonstrate that ClusterMI is more capable of identifying high-order SNP interactions from genome-wide data.

DualWMDR: Detecting epistatic interaction with dual screening and multifactor dimensionality reduction

“…For the three‐locus study, we do not take MOMDR as a comparing method for its overwhelming computational burden. Instead, we add two specifically designed high‐order epistasis approaches (ClusterMI, Cao et al, ; HiSeeker, Liu et al, ) for experiments. For a fair comparison, ClusterMI and HiSeeker employ the exhaustive search to detect epistatic interactions in the search stage.…”

“…Filter and wrapper based feature selection methodologies are two common screening methods (Long, Gianola, Rosa, Weigel, & Avendao, 2007;Saeys, Inza, & Larrañaga, 2007). Filter-based methods aim to select a subset of SNPs as a candidate set for interaction tests on the basis of existing biological knowledge (i.e., databases of pathways and protein-protein interactions; Ritchie, 2011;Turner et al, 2011), statistical features (i.e., marginal effects; Ma et al, 2012; and genotype frequencies; Ackermann & Beyer, 2012;Guo, Meng, Yu, & Pan, 2014;Xie et al, 2011) or fast algorithms (Cao, Yu, Liu, Jia, & Wang, 2018;Liu, Yu, Jiang, & Wang, 2017;Yang et al, 2008). Wrapper-based methods apply random sampling procedures (i.e., Markov chain Monte Carlo, MCMC; Zhang & Liu, 2007) and the Gibbs sampling; Tang, Wu, Jiang, & Li, 2009), heuristic algorithms (i.e., ant colony optimization, ACO; Sapin, Keedwell, & Frayling, 2015;Wang, Liu, Robbins, & Rekaya, 2010) and differential evolution, DE; C.-H. Yang et al, 2017) or machine learning algorithm (i.e., random forest, RF; Schwarz, König, & Ziegler, 2010;Yoshida & Koike, 2011), support vector machine (SVM; Chen et al, 2008;Marvel & Motsinger-Reif, 2012) and neural network (NN; Uppu et al, 2016) to search the space of interactions.…”

“…EDCF (Xie et al, 2011), DCHE (Guo et al, 2014), DECMDR (C.-H. Yang et al, 2017), HiSeeker (Liu et al, 2017), MOMDR (Yang et al, 2018), and ClusterMI (Cao et al, 2018). In addition, DualWMDR is evaluated on three real datasets, the age-related macular degeneration (AMD) data set (Klein et al, 2005), breast cancer (BC), and celiac disease (CD) datasets from the Wellcome Trust Case Control Consortium (WTCCC;…”

A Survey on Swarm Intelligence Search Methods Dedicated to Detection of High-Order SNP Interactions

“…An effective objective function has the ability to guide the SIS algorithm to explore some clues (such as different distributions of genotypes) that can further lead the algorithm to find high-order SNP interactions on a genome-wide scale. In existing research, the most common evaluation criteria (objective functions) involve the Bayesian-network-based score [58]- [60], mutual information [61], [62], logistic-regression-based score [63], [64], MDR [65], [66], Gini index [67]- [70], statistical test methods (e.g., chi-square test, G-test, and t-test) etc. These criteria usually have a high precision in evaluating a pure k-order SNP interaction (in which the k SNPs jointly affect complex diseases, and the number of SNPs is the same); however, they are ineffective for determining the association difference of SNP combinations that contain only some of the disease-causing SNPs.…”

Toxo: a library for calculating penetrance tables of high-order epistasis models

“…Furthermore, the expressions of ∈ (1, ∞). Recent studies that include simulations based on epistasis models to generate their evaluation data [20][21][22] settle on low-order models whose heritability values are worryingly moderate. However, real-world diseases are usually determined by a higher number of genes [1] and a higher heritability [23,24].…”