Wrapper-based selection of genetic features in genome-wide association studies through fast matrix operations

ABSTRACT: BACKGROUND: Through the wealth of information contained within them, genome-wide association studies (GWAS) have the potential to provide researchers with a systematic means of associating genetic variants with a wide variety of disease phenotypes. Due to the limitations of approaches that have analyzed single variants one at a time, it has been proposed that the genetic basis of these disorders could be determined through detailed analysis of the genetic variants themselves and in conjunction with one another. The construction of models that account for these subsets of variants requires methodologies that generate predictions based on the total risk of a particular group of polymorphisms. However, due to the excessive number of variants, constructing these types of models has so far been computationally infeasible. RESULTS: We have implemented an algorithm, known as greedy RLS, that we use to perform the first known wrapper-based feature selection on the genome-wide level. The running time of greedy RLS grows linearly in the number of training examples, the number of features in the original data set, and the number of selected features. This speed is achieved through computational short-cuts based on matrix calculus. Since the memory consumption in present-day computers can form an even tighter bottleneck than running time, we also developed a space efficient variation of greedy RLS which trades running time for memory. These approaches are then compared to traditional wrapper-based feature selection implementations based on support vector machines (SVM) to reveal the relative speed-up and to assess the feasibility of the new algorithm. As a proof of concept, we apply greedy RLS to the Hypertension - UK National Blood Service WTCCC dataset and select the most predictive variants using 3-fold external cross-validation in less than 26 minutes on a high end desktop. On this dataset, we also show that greedy RLS has a better classification performance on independent test data than a classifier trained using features selected by a statistical p-value-based filter, which is currently the most popular approach for constructing predictive models in GWAS. CONCLUSIONS: Greedy RLS is the first known implementation of a machine learning based method with the capability to conduct a wrapper-based feature selection on an entire GWAS containing several thousand examples and over 400,000 variants. In our experiments, greedy RLS selected a highly predictive subset of genetic variants in a fraction of the time spent by wrapper-based selection methods used together with SVM classifiers. The proposed algorithms are freely available as part of the RLScore software library at http://users.utu.fi/aatapa/RLScore/.     

Genetic control of canine leishmaniasis: genome-wide association study and genomic selection analysis

Background

The current disease model for leishmaniasis suggests that only a proportion of infected individuals develop clinical disease, while others are asymptomatically infected due to immune control of infection. The factors that determine whether individuals progress to clinical disease following Leishmania infection are unclear, although previous studies suggest a role for host genetics. Our hypothesis was that canine leishmaniasis is a complex disease with multiple loci responsible for the progression of the disease from Leishmania infection.

Methodology/Principal Findings

Genome-wide association and genomic selection approaches were applied to a population-based case-control dataset of 219 dogs from a single breed (Boxer) genotyped for ∼170,000 SNPs. Firstly, we aimed to identify individual disease loci; secondly, we quantified the genetic component of the observed phenotypic variance; and thirdly, we tested whether genome-wide SNP data could accurately predict the disease.

Conclusions/Significance

We estimated that a substantial proportion of the genome is affecting the trait and that its heritability could be as high as 60%. Using the genome-wide association approach, the strongest associations were on chromosomes 1, 4 and 20, although none of these were statistically significant at a genome-wide level and after correcting for genetic stratification and lifestyle. Amongst these associations, chromosome 4: 61.2–76.9 Mb maps to a locus that has previously been associated with host susceptibility to human and murine leishmaniasis, and genomic selection estimated markers in this region to have the greatest effect on the phenotype. We therefore propose these regions as candidates for replication studies. An important finding of this study was the significant predictive value from using the genomic information. We found that the phenotype could be predicted with an accuracy of ∼0.29 in new samples and that the affection status was correctly predicted in 60% of dogs, significantly higher than expected by chance, and with satisfactory sensitivity-specificity values (AUC = 0.63).     

Analysing biological pathways in genome-wide association studies

Genome-wide association (GWA) studies have typically focused on the analysis of single markers, which often lacks the power to uncover the relatively small effect sizes conferred by most genetic variants. Recently, pathway-based approaches have been developed, which use prior biological knowledge on gene function to facilitate more powerful analysis of GWA study data sets. These approaches typically examine whether a group of related genes in the same functional pathway are jointly associated with a trait of interest. Here we review the development of pathway-based approaches for GWA studies, discuss their practical use and caveats, and suggest that pathway-based approaches may also be useful for future GWA studies with sequencing data.     

Gene-centric characteristics of genome-wide association studies

Background

The high-throughput genotyping chips have contributed greatly to genome-wide association (GWA) studies to identify novel disease susceptibility single nucleotide polymorphisms (SNPs). The high-density chips are designed using two different SNP selection approaches, the direct gene-centric approach, and the indirect quasi-random SNPs or linkage disequilibrium (LD)-based tagSNPs approaches. Although all these approaches can provide high genome coverage and ascertain variants in genes, it is not clear to which extent these approaches could capture the common genic variants. It is also important to characterize and compare the differences between these approaches.

Methodology/Principal Findings

In our study, by using both the Phase II HapMap data and the disease variants extracted from OMIM, a gene-centric evaluation was first performed to evaluate the ability of the approaches in capturing the disease variants in Caucasian population. Then the distribution patterns of SNPs were also characterized in genic regions, evolutionarily conserved introns and nongenic regions, ontologies and pathways. The results show that, no mater which SNP selection approach is used, the current high-density SNP chips provide very high coverage in genic regions and can capture most of known common disease variants under HapMap frame. The results also show that the differences between the direct and the indirect approaches are relatively small. Both have similar SNP distribution patterns in these gene-centric characteristics.

Conclusions/Significance

This study suggests that the indirect approaches not only have the advantage of high coverage but also are useful for studies focusing on various functional SNPs either in genes or in the conserved regions that the direct approach supports. The study and the annotation of characteristics will be helpful for designing and analyzing GWA studies that aim to identify genetic risk factors involved in common diseases, especially variants in genes and conserved regions.     

Biostatistical aspects of genome-wide association studies

To search the entire human genome for association is a novel and promising approach to unravelling the genetic basis of complex genetic diseases. In these genome-wide association studies (GWAs), several hundreds of thousands of single nucleotide polymorphisms (SNPs) are analyzed at the same time, posing substantial biostatistical and computational challenges. In this paper, we discuss a number of biostatistical aspects of GWAs in detail. We specifically consider quality control issues and show that signal intensity plots are a sine qua condition non in today's GWAs. Approaches to detect and adjust for population stratification are briefly examined. We discuss different strategies aimed at tackling the problem of multiple testing, including adjustment of p -values, the false positive report probability and the false discovery rate. Another aspect of GWAs requiring special attention is the search for gene-gene and gene-environment interactions. We finally describe multistage approaches to GWAs.     

The road to genome-wide association studies

The recent crop of results from genome-wide association studies might seem like a sudden development. However, this blooming follows a long germination period during which the necessary concepts, resources and techniques were developed and assembled. Here, I look back at how the necessary pieces fell into place, focusing on the less well-chronicled days before the launch of the HapMap project, and speculate about future developments.     

The meta-analysis of genome-wide association studies

The pressure to publish novel genetic associations has meant that meta-analysis has been applied to genome-wide association studies without the time for a careful consideration of the methods that are used. This review distinguishes between the use of meta-analysis to validate previously reported genetic associations and its use for gene discovery, and advocates viewing gene discovery as an exploratory screen that requires independent replication instead of treating it as the application of hundreds of thousands of statistical tests. The review considers the use of fixed and random effects meta-analyses, the investigation of between-study heterogeneity, adjustment for confounding, assessing the combined evidence and genomic control, and comments on alternative approaches that have been used in the literature.     

Family-based designs for genome-wide association studies

Association mapping has successfully identified common SNPs associated with many diseases. However, the inability of this class of variation to account for most of the supposed heritability has led to a renewed interest in methods - primarily linkage analysis - to detect rare variants. Family designs allow for control of population stratification, investigations of questions such as parent-of-origin effects and other applications that are imperfectly or not readily addressed in case-control association studies. This article guides readers through the interface between linkage and association analysis, reviews the new methodologies and provides useful guidelines for applications. Just as effective SNP-genotyping tools helped to realize the potential of association studies, next-generation sequencing tools will benefit genetic studies by improving the power of family-based approaches.     

Interpreting meta-analyses of genome-wide association studies

Meta-analysis is an increasingly popular tool for combining multiple genome-wide association studies in a single analysis to identify associations with small effect sizes. The effect sizes between studies in a meta-analysis may differ and these differences, or heterogeneity, can be caused by many factors. If heterogeneity is observed in the results of a meta-analysis, interpreting the cause of heterogeneity is important because the correct interpretation can lead to a better understanding of the disease and a more effective design of a replication study. However, interpreting heterogeneous results is difficult. The standard approach of examining the association p-values of the studies does not effectively predict if the effect exists in each study. In this paper, we propose a framework facilitating the interpretation of the results of a meta-analysis. Our framework is based on a new statistic representing the posterior probability that the effect exists in each study, which is estimated utilizing cross-study information. Simulations and application to the real data show that our framework can effectively segregate the studies predicted to have an effect, the studies predicted to not have an effect, and the ambiguous studies that are underpowered. In addition to helping interpretation, the new framework also allows us to develop a new association testing procedure taking into account the existence of effect.     

Genotyping and inflated type I error rate in genome-wide association case/control studies

Background  

One common goal of a case/control genome wide association study (GWAS) is to find SNPs associated with a disease. Traditionally, the first step in such studies is to assign a genotype to each SNP in each subject, based on a statistic summarizing fluorescence measurements. When the distributions of the summary statistics are not well separated by genotype, the act of genotype assignment can lead to more potential problems than acknowledged by the literature.     

Genomic consequences of selection and genome-wide association mapping in soybean

Background

Crop improvement always involves selection of specific alleles at genes controlling traits of agronomic importance, likely resulting in detectable signatures of selection within the genome of modern soybean (Glycine max L. Merr.). The identification of these signatures of selection is meaningful from the perspective of evolutionary biology and for uncovering the genetic architecture of agronomic traits.

Results

To this end, two populations of soybean, consisting of 342 landraces and 1062 improved lines, were genotyped with the SoySNP50K Illumina BeadChip containing 52,041 single nucleotide polymorphisms (SNPs), and systematically phenotyped for 9 agronomic traits. A cross-population composite likelihood ratio (XP-CLR) method was used to screen the signals of selective sweeps. A total of 125 candidate selection regions were identified, many of which harbored genes potentially involved in crop improvement. To further investigate whether these candidate regions were in fact enriched for genes affected by selection, genome-wide association studies (GWAS) were conducted on 7 selection traits targeted in soybean breeding (grain yield, plant height, lodging, maturity date, seed coat color, seed protein and oil content) and 2 non-selection traits (pubescence and flower color). Major genomic regions associated with selection traits overlapped with candidate selection regions, whereas no overlap of this kind occurred for the non-selection traits, suggesting that the selection sweeps identified are associated with traits of agronomic importance. Multiple novel loci and refined map locations of known loci related to these traits were also identified.

Conclusions

These findings illustrate that comparative genomic analyses, especially when combined with GWAS, are a promising approach to dissect the genetic architecture of complex traits.

Electronic supplementary material

The online version of this article (doi:10.1186/s12864-015-1872-y) contains supplementary material, which is available to authorized users.     

A robust method for testing association in genome-wide association studies

In genetic association studies, due to the varying underlying genetic models, no single statistical test can be the most powerful test under all situations. Current studies show that if the underlying genetic models are known, trend-based tests, which outperform the classical Pearson χ2 test, can be constructed. However, when the underlying genetic models are unknown, the χ2 test is usually more robust than trend-based tests. In this paper, we propose a new association test based on a generalized genetic model, namely the generalized order-restricted relative risks model. Through a Monte Carlo simulation study, we show that the proposed association test is generally more powerful than the χ2 test, and more robust than those trend-based tests. The proposed methodologies are also illustrated by some real SNP datasets.     

Designing candidate gene and genome-wide case-control association studies

This protocol describes how to appropriately design a genetic association case-control study, either focusing on a candidate gene (CG) or region or implementing a genome-wide approach. The steps described involve: (i) defining the case phenotype in adequate detail; (ii) checking the heritability of the disease in question; (iii) considering whether a population-based study is the appropriate design for the research question; (iv) the appropriate selection of controls; (v) sample size calculations and (vi) giving due consideration to whether it is a de novo or replication study. General guidelines are given, as well as specific examples of a CG and a genome-wide association study into type 2 diabetes. Software and websites used in this protocol include the International HapMap Consortium website, Genetic Power Calculator, CaT, and SNPSpD. Running each of the programs takes only a few seconds; the rate-limiting steps involve thinking through the designs and parameters in the disease models.     

Current status of genome-wide association studies in cancer

Genome-wide association studies in cancer have already identified over 150 regions associated with two dozen specific cancers. Already, a handful of multi-cancer susceptibility regions have been uncovered, providing new insights into perhaps common mechanisms of carcinogenesis. For each new susceptibility allele, investigators now face the arduous task of interrogating each region beginning with fine mapping prior to pursuing the biological basis for the direct association of one or more variants. It appears that there may be a significant number of common alleles that contribute to the heritability of a specific cancer. Since each region confers a small contribution to the risk for cancer, it is daunting to consider any single nucleotide polymorphism (SNP) as a clinical test. Since the complex genomic architecture of each cancer differs, additional genotyping and sequence analysis will be required to comprehensively catalog susceptibility alleles followed by the formidable task of understanding the interactions between genetic regions as well as the environment. It will be critical to assess the applicability of genetic tests in specific clinical settings, such as when to perform screening tests with calculable risks (e.g., biopsies or chemoprevention), before incorporating SNPs into clinical practice. To advance the current genomic observations to the clinical venue, new studies will need to be designed to validate the utility of known genetic variants in assessing risk for cancer as well as its outcomes.