Genomic similarity and kernel methods I: advancements by building on mathematical and statistical foundations

A kernel regression approach to gene-gene interaction detection for case-control studies

Gene-gene interactions are increasingly being addressed as a potentially important contributor to the variability of complex traits. Consequently, attentions have moved beyond single locus analysis of association to more complex genetic models. Although several single-marker approaches toward interaction analysis have been developed, such methods suffer from very high testing dimensionality and do not take advantage of existing information, notably the definition of genes as functional units. Here, we propose a comprehensive family of gene-level score tests for identifying genetic elements of disease risk, in particular pairwise gene-gene interactions. Using kernel machine methods, we devise score-based variance component tests under a generalized linear mixed model framework. We conducted simulations based upon coalescent genetic models to evaluate the performance of our approach under a variety of disease models. These simulations indicate that our methods are generally higher powered than alternative gene-level approaches and at worst competitive with exhaustive SNP-level (where SNP is single-nucleotide polymorphism) analyses. Furthermore, we observe that simulated epistatic effects resulted in significant marginal testing results for the involved genes regardless of whether or not true main effects were present. We detail the benefits of our methods and discuss potential genome-wide analysis strategies for gene-gene interaction analysis in a case-control study design.

SBERIA: set-based gene-environment interaction test for rare and common variants in complex diseases

Identification of gene-environment interaction (G × E) is important in understanding the etiology of complex diseases. However, partially due to the lack of power, there have been very few replicated G × E findings compared to the success in marginal association studies. The existing G × E testing methods mainly focus on improving the power for individual markers. In this paper, we took a different strategy and proposed a set-based gene-environment interaction test (SBERIA), which can improve the power by reducing the multiple testing burdens and aggregating signals within a set. The major challenge of the signal aggregation within a set is how to tell signals from noise and how to determine the direction of the signals. SBERIA takes advantage of the established correlation screening for G × E to guide the aggregation of genotypes within a marker set. The correlation screening has been shown to be an efficient way of selecting potential G × E candidate SNPs in case-control studies for complex diseases. Importantly, the correlation screening in case-control combined samples is independent of the interaction test. With this desirable feature, SBERIA maintains the correct type I error level and can be easily implemented in a regular logistic regression setting. We showed that SBERIA had higher power than benchmark methods in various simulation scenarios, both for common and rare variants. We also applied SBERIA to real genome-wide association studies (GWAS) data of 10,729 colorectal cancer cases and 13,328 controls and found evidence of interaction between the set of known colorectal cancer susceptibility loci and smoking.

Nonlinear dimension reduction with Wright-Fisher kernel for genotype aggregation and association mapping

MOTIVATION

Association tests based on next-generation sequencing data are often under-powered due to the presence of rare variants and large amount of neutral or protective variants. A successful strategy is to aggregate genetic information within meaningful single-nucleotide polymorphism (SNP) sets, e.g. genes or pathways, and test association on SNP sets. Many existing methods for group-wise tests require specific assumptions about the direction of individual SNP effects and/or perform poorly in the presence of interactions.

RESULTS

We propose a joint association test strategy based on two key components: a nonlinear supervised dimension reduction approach for effective SNP information aggregation and a novel kernel specially designed for qualitative genotype data. The new test demonstrates superior performance in identifying causal genes over existing methods across a large variety of disease models simulated from sequence data of real genes. In general, the proposed method provides an association test strategy that can (i) detect both rare and common causal variants, (ii) deal with both additive and interaction effect, (iii) handle both quantitative traits and disease dichotomies and (iv) incorporate non-genetic covariates. In addition, the new kernel can potentially boost the power of the entire family of kernel-based methods for genetic data analysis.

AVAILABILITY

The method is implemented in MATLAB. Source code is available upon request.

CONTACT

hongjie.zhu@duke.edu.

Multiple genetic variant association testing by collapsing and kernel methods with pedigree or population structured data

Searching for rare genetic variants associated with complex diseases can be facilitated by enriching for diseased carriers of rare variants by sampling cases from pedigrees enriched for disease, possibly with related or unrelated controls. This strategy, however, complicates analyses because of shared genetic ancestry, as well as linkage disequilibrium among genetic markers. To overcome these problems, we developed broad classes of "burden" statistics and kernel statistics, extending commonly used methods for unrelated case-control data to allow for known pedigree relationships, for autosomes and the X chromosome. Furthermore, by replacing pedigree-based genetic correlation matrices with estimates of genetic relationships based on large-scale genomic data, our methods can be used to account for population-structured data. By simulations, we show that the type I error rates of our developed methods are near the asymptotic nominal levels, allowing rapid computation of P-values. Our simulations also show that a linear weighted kernel statistic is generally more powerful than a weighted "burden" statistic. Because the proposed statistics are rapid to compute, they can be readily used for large-scale screening of the association of genomic sequence data with disease status.

Kernel machine SNP-set testing under multiple candidate kernels

Joint testing for the cumulative effect of multiple single-nucleotide polymorphisms grouped on the basis of prior biological knowledge has become a popular and powerful strategy for the analysis of large-scale genetic association studies. The kernel machine (KM)-testing framework is a useful approach that has been proposed for testing associations between multiple genetic variants and many different types of complex traits by comparing pairwise similarity in phenotype between subjects to pairwise similarity in genotype, with similarity in genotype defined via a kernel function. An advantage of the KM framework is its flexibility: choosing different kernel functions allows for different assumptions concerning the underlying model and can allow for improved power. In practice, it is difficult to know which kernel to use a priori because this depends on the unknown underlying trait architecture and selecting the kernel which gives the lowest P-value can lead to inflated type I error. Therefore, we propose practical strategies for KM testing when multiple candidate kernels are present based on constructing composite kernels and based on efficient perturbation procedures. We demonstrate through simulations and real data applications that the procedures protect the type I error rate and can lead to substantially improved power over poor choices of kernels and only modest differences in power vs. using the best candidate kernel.

A novel kernel for correcting size bias in the logistic kernel machine test with an application to rheumatoid arthritis

OBJECTIVES

The logistic kernel machine test (LKMT) is a testing procedure tailored towards high-dimensional genetic data. Its use in pathway analyses of case-control genome-wide association studies results from its computational efficiency and flexibility in incorporating additional information via the kernel. The kernel can be any positive definite function; unfortunately, its form strongly influences the test's power and bias. Most authors have recommended the use of a simple linear kernel. We demonstrate via a simulation that the probability of rejecting the null hypothesis of no association just by chance increases with the number of SNPs or genes in the pathway when applying a simple linear kernel.

METHODS

We propose a novel kernel that includes an appropriate standardization in order to protect against any inflation of false positive results. Moreover, our novel kernel contains information on gene membership of SNPs in the pathway.

RESULTS

When applying the novel kernel to data from the North American Rheumatoid Arthritis Consortium, we find that even this basic genomic structure can improve the ability of the LKMT to identify meaningful associations. We also demonstrate that the standardization effectively eliminates problems of size bias.

CONCLUSION

We recommend the use of our standardized kernel and urge caution when using non-adjusted kernels in the LKMT to conduct pathway analyses.

Comparison of similarity-based tests and pooling strategies for rare variants

Background

As several rare genomic variants have been shown to affect common phenotypes, rare variants association analysis has received considerable attention. Several efficient association tests using genotype and phenotype similarity measures have been proposed in the literature. The major advantages of similarity-based tests are their ability to accommodate multiple types of DNA variations within one association test, and to account for the possible interaction within a region. However, not much work has been done to compare the performance of similarity-based tests on rare variants association scenarios, especially when applied with different rare variants pooling strategies.

Results

Based on the population genetics simulations and analysis of a publicly-available sequencing data set, we compared the performance of four similarity-based tests and two rare variants pooling strategies. We showed that weighting approach outperforms collapsing under the presence of strong effect from rare variants and under the presence of moderate effect from common variants, whereas collapsing of rare variants is preferable when common variants possess a strong effect. We also demonstrated that the difference in statistical power between the two pooling strategies may be substantial. The results also highlighted consistently high power of two similarity-based approaches when applied with an appropriate pooling strategy.

Conclusions

Population genetics simulations and sequencing data set analysis showed high power of two similarity-based tests and a substantial difference in power between the two pooling strategies.

Increasing power for voxel-wise genome-wide association studies: the random field theory, least square kernel machines and fast permutation procedures

Imaging traits are thought to have more direct links to genetic variation than diagnostic measures based on cognitive or clinical assessments and provide a powerful substrate to examine the influence of genetics on human brains. Although imaging genetics has attracted growing attention and interest, most brain-wide genome-wide association studies focus on voxel-wise single-locus approaches, without taking advantage of the spatial information in images or combining the effect of multiple genetic variants. In this paper we present a fast implementation of voxel- and cluster-wise inferences based on the random field theory to fully use the spatial information in images. The approach is combined with a multi-locus model based on least square kernel machines to associate the joint effect of several single nucleotide polymorphisms (SNP) with imaging traits. A fast permutation procedure is also proposed which significantly reduces the number of permutations needed relative to the standard empirical method and provides accurate small p-value estimates based on parametric tail approximation. We explored the relation between 448,294 single nucleotide polymorphisms and 18,043 genes in 31,662 voxels of the entire brain across 740 elderly subjects from the Alzheimer's disease neuroimaging initiative (ADNI). Structural MRI scans were analyzed using tensor-based morphometry (TBM) to compute 3D maps of regional brain volume differences compared to an average template image based on healthy elderly subjects. We find method to be more sensitive compared with voxel-wise single-locus approaches. A number of genes were identified as having significant associations with volumetric changes. The most associated gene was GRIN2B, which encodes the N-methyl-d-aspartate (NMDA) glutamate receptor NR2B subunit and affects both the parietal and temporal lobes in human brains. Its role in Alzheimer's disease has been widely acknowledged and studied, suggesting the validity of the approach. The various advantages over existing approaches indicate a great potential offered by this novel framework to detect genetic influences on human brains.

Network methods for describing sample relationships in genomic datasets: application to Huntington's disease

BACKGROUND

Genomic datasets generated by new technologies are increasingly prevalent in disparate areas of biological research. While many studies have sought to characterize relationships among genomic features, commensurate efforts to characterize relationships among biological samples have been less common. Consequently, the full extent of sample variation in genomic studies is often under-appreciated, complicating downstream analytical tasks such as gene co-expression network analysis.

RESULTS

Here we demonstrate the use of network methods for characterizing sample relationships in microarray data generated from human brain tissue. We describe an approach for identifying outlying samples that does not depend on the choice or use of clustering algorithms. We introduce a battery of measures for quantifying the consistency and integrity of sample relationships, which can be compared across disparate studies, technology platforms, and biological systems. Among these measures, we provide evidence that the correlation between the connectivity and the clustering coefficient (two important network concepts) is a sensitive indicator of homogeneity among biological samples. We also show that this measure, which we refer to as cor(K,C), can distinguish biologically meaningful relationships among subgroups of samples. Specifically, we find that cor(K,C) reveals the profound effect of Huntington's disease on samples from the caudate nucleus relative to other brain regions. Furthermore, we find that this effect is concentrated in specific modules of genes that are naturally co-expressed in human caudate nucleus, highlighting a new strategy for exploring the effects of disease on sets of genes.

CONCLUSIONS

These results underscore the importance of systematically exploring sample relationships in large genomic datasets before seeking to analyze genomic feature activity. We introduce a standardized platform for this purpose using freely available R software that has been designed to enable iterative and interactive exploration of sample networks.

Model selection and psychological theory: a discussion of the differences between the Akaike information criterion (AIC) and the Bayesian information criterion (BIC)

This article reviews the Akaike information criterion (AIC) and the Bayesian information criterion (BIC) in model selection and the appraisal of psychological theory. The focus is on latent variable models, given their growing use in theory testing and construction. Theoretical statistical results in regression are discussed, and more important issues are illustrated with novel simulations involving latent variable models including factor analysis, latent profile analysis, and factor mixture models. Asymptotically, the BIC is consistent, in that it will select the true model if, among other assumptions, the true model is among the candidate models considered. The AIC is not consistent under these circumstances. When the true model is not in the candidate model set the AIC is efficient, in that it will asymptotically choose whichever model minimizes the mean squared error of prediction/estimation. The BIC is not efficient under these circumstances. Unlike the BIC, the AIC also has a minimax property, in that it can minimize the maximum possible risk in finite sample sizes. In sum, the AIC and BIC have quite different properties that require different assumptions, and applied researchers and methodologists alike will benefit from improved understanding of the asymptotic and finite-sample behavior of these criteria. The ultimate decision to use the AIC or BIC depends on many factors, including the loss function employed, the study's methodological design, the substantive research question, and the notion of a true model and its applicability to the study at hand.

On the Aggregation of Multimarker Information for Marker-Set and Sequencing Data Analysis: Genotype Collapsing vs. Similarity Collapsing

Methods that collapse information across genetic markers when searching for association signals are gaining momentum in the literature. Although originally developed to achieve a better balance between retaining information and controlling degrees of freedom when performing multimarker association analysis, these methods have recently been proven to be a powerful tool for identifying rare variants that contribute to complex phenotypes. The information among markers can be collapsed at the genotype level, which focuses on the mean of genetic information, or the similarity level, which focuses on the variance of genetic information. The aim of this work is to understand the strengths and weaknesses of these two collapsing strategies. Our results show that neither collapsing strategy outperforms the other across all simulated scenarios. Two factors that dominate the performance of these strategies are the signal-to-noise ratio and the underlying genetic architecture of the causal variants. Genotype collapsing is more sensitive to the marker set being contaminated by noise loci than similarity collapsing. In addition, genotype collapsing performs best when the genetic architecture of the causal variants is not complex (e.g., causal loci with similar effects and similar frequencies). Similarity collapsing is more robust as the complexity of the genetic architecture increases and outperforms genotype collapsing when the genetic architecture of the marker set becomes more sophisticated (e.g., causal loci with various effect sizes or frequencies and potential non-linear or interactive effects). Because the underlying genetic architecture is not known a priori, we also considered a two-stage analysis that combines the two top-performing methods from different collapsing strategies. We find that it is reasonably robust across all simulated scenarios.

Testing gene-environment interactions in gene-based association studies

Gene-based and single-nucleotide polymorphism (SNP) set association studies provide an important complement to SNP analysis. Kernel-based nonparametric regression has recently emerged as a powerful and flexible tool for this purpose. Our goal is to explore whether this approach can be extended to incorporate and test for interaction effects, especially for genes containing rare variant SNPs. Here, we construct nonparametric regression models that can be used to include a gene-environment interaction effect under the framework of the least-squares kernel machine and examine the performance of the proposed method on the Genetic Analysis Workshop 17 unrelated individuals data set. Two hundred simulated replicates were used to explore the power for detecting interaction. We demonstrate through a genome scan of the quantitative phenotype Q1 that the simulated gene-environment interaction effect in the data can be detected with reasonable power by using the least-squares kernel machine method.

Distance-based phenotypic association analysis of DNA sequence data

As the cost of sequencing decreases, the demand for association tests that use exhaustive DNA sequence information increases. One such association test is multivariate distance matrix regression (MDMR). We explore some of the features of MDMR using Genetic Analysis Workshop 17 simulated data in search of potential improvements in distance measures. We used genotype data from 697 unrelated individuals, in 200 replications, to test the power of MDMR to detect 13 trait Q2 causative genes based on the Euclidean distance metric. We also estimated the false-positive rate of MDMR using 508 control genes. In addition, we compared MDMR with Mantel's test and collapsing analysis for rare variants. MDMR performed comparably well even with the Euclidean distance measure.

Multilocus association testing with penalized regression

In multilocus association analysis, since some markers may not be associated with a trait, it seems attractive to use penalized regression with the capability of automatic variable selection. On the other hand, in spite of a rapidly growing body of literature on penalized regression, most focus on variable selection and outcome prediction, for which penalized methods are generally more effective than their nonpenalized counterparts. However, for statistical inference, i.e. hypothesis testing and interval estimation, it is less clear how penalized methods would perform, or even how to best apply them, largely due to lack of studies on this topic. In our motivating data for a cohort of kidney transplant recipients, it is of primary interest to assess whether a group of genetic variants are associated with a binary clinical outcome, acute rejection at 6 months. In this article, we study some technical issues and alternative implementations of hypothesis testing in Lasso penalized logistic regression, and compare their performance with each other and with several existing global tests, some of which are specifically designed as variance component tests for high-dimensional data. The most interesting, and perhaps surprising, conclusion of this study is that, for low to moderately high-dimensional data, statistical tests based on Lasso penalized regression are not necessarily more powerful than some existing global tests. In addition, in penalized regression, rather than building a test based on a single selected "best" model, combining multiple tests, each of which is built on a candidate model, might be more promising.

Studying gene and gene-environment effects of uncommon and common variants on continuous traits: a marker-set approach using gene-trait similarity regression

Genomic association analyses of complex traits demand statistical tools that are capable of detecting small effects of common and rare variants and modeling complex interaction effects and yet are computationally feasible. In this work, we introduce a similarity-based regression method for assessing the main genetic and interaction effects of a group of markers on quantitative traits. The method uses genetic similarity to aggregate information from multiple polymorphic sites and integrates adaptive weights that depend on allele frequencies to accomodate common and uncommon variants. Collapsing information at the similarity level instead of the genotype level avoids canceling signals that have the opposite etiological effects and is applicable to any class of genetic variants without the need for dichotomizing the allele types. To assess gene-trait associations, we regress trait similarities for pairs of unrelated individuals on their genetic similarities and assess association by using a score test whose limiting distribution is derived in this work. The proposed regression framework allows for covariates, has the capacity to model both main and interaction effects, can be applied to a mixture of different polymorphism types, and is computationally efficient. These features make it an ideal tool for evaluating associations between phenotype and marker sets defined by linkage disequilibrium (LD) blocks, genes, or pathways in whole-genome analysis.

Comparison of statistical tests for disease association with rare variants

In anticipation of the availability of next-generation sequencing data, there is increasing interest in investigating association between complex traits and rare variants (RVs). In contrast to association studies for common variants (CVs), due to the low frequencies of RVs, common wisdom suggests that existing statistical tests for CVs might not work, motivating the recent development of several new tests for analyzing RVs, most of which are based on the idea of pooling/collapsing RVs. However, there is a lack of evaluations of, and thus guidance on the use of, existing tests. Here we provide a comprehensive comparison of various statistical tests using simulated data. We consider both independent and correlated rare mutations, and representative tests for both CVs and RVs. As expected, if there are no or few non-causal (i.e. neutral or non-associated) RVs in a locus of interest while the effects of causal RVs on the trait are all (or mostly) in the same direction (i.e. either protective or deleterious, but not both), then the simple pooled association tests (without selecting RVs and their association directions) and a new test called kernel-based adaptive clustering (KBAC) perform similarly and are most powerful; KBAC is more robust than simple pooled association tests in the presence of non-causal RVs; however, as the number of non-causal CVs increases and/or in the presence of opposite association directions, the winners are two methods originally proposed for CVs and a new test called C-alpha test proposed for RVs, each of which can be regarded as testing on a variance component in a random-effects model. Interestingly, several methods based on sequential model selection (i.e. selecting causal RVs and their association directions), including two new methods proposed here, perform robustly and often have statistical power between those of the above two classes.

Predicting genetic values: a kernel-based best linear unbiased prediction with genomic data

Genomic data provide a valuable source of information for modeling covariance structures, allowing a more accurate prediction of total genetic values (GVs). We apply the kriging concept, originally developed in the geostatistical context for predictions in the low-dimensional space, to the high-dimensional space spanned by genomic single nucleotide polymorphism (SNP) vectors and study its properties in different gene-action scenarios. Two different kriging methods [“universal kriging” (UK) and “simple kriging” (SK)] are presented. As a novelty, we suggest use of the family of Matérn covariance functions to model the covariance structure of SNP vectors. A genomic best linear unbiased prediction (GBLUP) is applied as a reference method. The three approaches are compared in a whole-genome simulation study considering additive, additive-dominance, and epistatic gene-action models. Predictive performance is measured in terms of correlation between true and predicted GVs and average true GVs of the individuals ranked best by prediction. We show that UK outperforms GBLUP in the presence of dominance and epistatic effects. In a limiting case, it is shown that the genomic covariance structure proposed by VanRaden (2008) can be considered as a covariance function with corresponding quadratic variogram. We also prove theoretically that if a specific linear relationship exists between covariance matrices for two linear mixed models, the GVs resulting from BLUP are linked by a scaling factor. Finally, the relation of kriging to other models is discussed and further options for modeling the covariance structure, which might be more appropriate in the genomic context, are suggested.

Genomic similarity and kernel methods II: methods for genomic information

Measures of genomic similarity are often the basis of flexible statistical analyses, and when based on kernel methods, they provide a powerful platform to take advantage of a broad and deep statistical theory, and a wide range of existing software; see the companion paper for a review of this material [1]. The kernel method converts information - perhaps complex or high-dimensional information - for a pair of subjects to a quantitative value representing either similarity or dissimilarity, with the requirement that it must create a positive semidefinite matrix when applied to all pairs of subjects. This approach provides enormous opportunities to enhance genetic analyses by including a wide range of publically-available data as structured kernel 'prior' information. Kernel methods are appealing for their generality, yet this generality can make it challenging to formulate measures of similarity that directly address a specific scientific aim, or that are most powerful to detect a specific genetic mechanism. Although it is difficult to create a cook book of kernels for genetic studies, useful guidelines can be gleaned from a variety of novel published approaches. We review some novel developments of kernels for specific analyses and speculate on how to build kernels for complex genomic attributes based on publically available data. The creativity of analysts, with rigorous evaluations by applications to real and simulated data, will ultimately provide a much stronger array of kernel 'tools' for genetic analyses.