Self-Contained Gene-Set Analysis of Expression Data: An Evaluation of Existing and Novel Methods

Gene set methods aim to assess the overall evidence of association of a set of genes with a phenotype, such as disease or a quantitative trait. Multiple approaches for gene set analysis of expression data have been proposed. They can be divided into two types: competitive and self-contained. Benefits of self-contained methods include that they can be used for genome-wide, candidate gene, or pathway studies, and have been reported to be more powerful than competitive methods. We therefore investigated ten self-contained methods that can be used for continuous, discrete and time-to-event phenotypes. To assess the power and type I error rate for the various previously proposed and novel approaches, an extensive simulation study was completed in which the scenarios varied according to: number of genes in a gene set, number of genes associated with the phenotype, effect sizes, correlation between expression of genes within a gene set, and the sample size. In addition to the simulated data, the various methods were applied to a pharmacogenomic study of the drug gemcitabine. Simulation results demonstrated that overall Fisher''s method and the global model with random effects have the highest power for a wide range of scenarios, while the analysis based on the first principal component and Kolmogorov-Smirnov test tended to have lowest power. The methods investigated here are likely to play an important role in identifying pathways that contribute to complex traits.     

PUMA: A Unified Framework for Penalized Multiple Regression Analysis of GWAS Data

Penalized Multiple Regression (PMR) can be used to discover novel disease associations in GWAS datasets. In practice, proposed PMR methods have not been able to identify well-supported associations in GWAS that are undetectable by standard association tests and thus these methods are not widely applied. Here, we present a combined algorithmic and heuristic framework for PUMA (Penalized Unified Multiple-locus Association) analysis that solves the problems of previously proposed methods including computational speed, poor performance on genome-scale simulated data, and identification of too many associations for real data to be biologically plausible. The framework includes a new minorize-maximization (MM) algorithm for generalized linear models (GLM) combined with heuristic model selection and testing methods for identification of robust associations. The PUMA framework implements the penalized maximum likelihood penalties previously proposed for GWAS analysis (i.e. Lasso, Adaptive Lasso, NEG, MCP), as well as a penalty that has not been previously applied to GWAS (i.e. LOG). Using simulations that closely mirror real GWAS data, we show that our framework has high performance and reliably increases power to detect weak associations, while existing PMR methods can perform worse than single marker testing in overall performance. To demonstrate the empirical value of PUMA, we analyzed GWAS data for type 1 diabetes, Crohns''s disease, and rheumatoid arthritis, three autoimmune diseases from the original Wellcome Trust Case Control Consortium. Our analysis replicates known associations for these diseases and we discover novel etiologically relevant susceptibility loci that are invisible to standard single marker tests, including six novel associations implicating genes involved in pancreatic function, insulin pathways and immune-cell function in type 1 diabetes; three novel associations implicating genes in pro- and anti-inflammatory pathways in Crohn''s disease; and one novel association implicating a gene involved in apoptosis pathways in rheumatoid arthritis. We provide software for applying our PUMA analysis framework.     

Incorporating regulatory interactions into gene-set analyses for GWAS data: A controlled analysis with the MAGMA tool

To date, genome-wide association studies have identified thousands of statistically-significant associations between genetic variants, and phenotypes related to a myriad of traits and diseases. A key goal for human-genetics research is to translate these associations into functional mechanisms. Popular gene-set analysis tools, like MAGMA, map variants to genes they might affect, and then integrate genome-wide association study data (that is, variant-level associations for a phenotype) to score genes for association with a phenotype. Gene scores are subsequently used in competitive gene-set analyses to identify biological processes that are enriched for phenotype association. By default, variants are mapped to genes in their proximity. However, many variants that affect phenotypes are thought to act at regulatory elements, which can be hundreds of kilobases away from their target genes. Thus, we explored the idea of augmenting a proximity-based mapping scheme with publicly-available datasets of regulatory interactions. We used MAGMA to analyze genome-wide association study data for ten different phenotypes, and evaluated the effects of augmentation by comparing numbers, and identities, of genes and gene sets detected as statistically significant between mappings. We detected several pitfalls and confounders of such “augmented analyses”, and introduced ways to control for them. Using these controls, we demonstrated that augmentation with datasets of regulatory interactions only occasionally strengthened the enrichment for phenotype association amongst (biologically-relevant) gene sets for different phenotypes. Still, in such cases, genes and regulatory elements responsible for the improvement could be pinpointed. For instance, using brain regulatory-interactions for augmentation, we were able to implicate two acetylcholine receptor subunits involved in post-synaptic chemical transmission, namely CHRNB2 and CHRNE, in schizophrenia. Collectively, our study presents a critical approach for integrating regulatory interactions into gene-set analyses for genome-wide association study data, by introducing various controls to distinguish genuine results from spurious discoveries.     

Accuracy of CNV Detection from GWAS Data

Several computer programs are available for detecting copy number variants (CNVs) using genome-wide SNP arrays. We evaluated the performance of four CNV detection software suites--Birdsuite, Partek, HelixTree, and PennCNV-Affy--in the identification of both rare and common CNVs. Each program's performance was assessed in two ways. The first was its recovery rate, i.e., its ability to call 893 CNVs previously identified in eight HapMap samples by paired-end sequencing of whole-genome fosmid clones, and 51,440 CNVs identified by array Comparative Genome Hybridization (aCGH) followed by validation procedures, in 90 HapMap CEU samples. The second evaluation was program performance calling rare and common CNVs in the Bipolar Genome Study (BiGS) data set (1001 bipolar cases and 1033 controls, all of European ancestry) as measured by the Affymetrix SNP 6.0 array. Accuracy in calling rare CNVs was assessed by positive predictive value, based on the proportion of rare CNVs validated by quantitative real-time PCR (qPCR), while accuracy in calling common CNVs was assessed by false positive/false negative rates based on qPCR validation results from a subset of common CNVs. Birdsuite recovered the highest percentages of known HapMap CNVs containing >20 markers in two reference CNV datasets. The recovery rate increased with decreased CNV frequency. In the tested rare CNV data, Birdsuite and Partek had higher positive predictive values than the other software suites. In a test of three common CNVs in the BiGS dataset, Birdsuite's call was 98.8% consistent with qPCR quantification in one CNV region, but the other two regions showed an unacceptable degree of accuracy. We found relatively poor consistency between the two "gold standards," the sequence data of Kidd et al., and aCGH data of Conrad et al. Algorithms for calling CNVs especially common ones need substantial improvement, and a "gold standard" for detection of CNVs remains to be established.     

Multi-SNP Analysis of GWAS Data Identifies Pathways Associated with Nonalcoholic Fatty Liver Disease

Non-alcoholic fatty liver disease (NAFLD) is a common liver disease; the histological spectrum of which ranges from steatosis to steatohepatitis. Nonalcoholic steatohepatitis (NASH) often leads to cirrhosis and development of hepatocellular carcinoma. To better understand pathogenesis of NAFLD, we performed the pathway of distinction analysis (PoDA) on a genome-wide association study dataset of 250 non-Hispanic white female adult patients with NAFLD, who were enrolled in the NASH Clinical Research Network (CRN) Database Study, to investigate whether biologic process variation measured through genomic variation of genes within these pathways was related to the development of steatohepatitis or cirrhosis. Pathways such as Recycling of eIF2:GDP, biosynthesis of steroids, Terpenoid biosynthesis and Cholesterol biosynthesis were found to be significantly associated with NASH. SNP variants in Terpenoid synthesis, Cholesterol biosynthesis and biosynthesis of steroids were associated with lobular inflammation and cytologic ballooning while those in Terpenoid synthesis were also associated with fibrosis and cirrhosis. These were also related to the NAFLD activity score (NAS) which is derived from the histological severity of steatosis, inflammation and ballooning degeneration. Eukaryotic protein translation and recycling of eIF2:GDP related SNP variants were associated with ballooning, steatohepatitis and cirrhosis. Il2 signaling events mediated by PI3K, Mitotic metaphase/anaphase transition, and Prostanoid ligand receptors were also significantly associated with cirrhosis. Taken together, the results provide evidence for additional ways, beyond the effects of single SNPs, by which genetic factors might contribute to the susceptibility to develop a particular phenotype of NAFLD and then progress to cirrhosis. Further studies are warranted to explain potential important genetic roles of these biological processes in NAFLD.     

Network-Based Multiple Sclerosis Pathway Analysis with GWAS Data from 15,000 Cases and 30,000 Controls

Multiple sclerosis (MS) is an inflammatory CNS disease with a substantial genetic component, originally mapped to only the human leukocyte antigen (HLA) region. In the last 5 years, a total of seven genome-wide association studies and one meta-analysis successfully identified 57 non-HLA susceptibility loci. Here, we merged nominal statistical evidence of association and physical evidence of interaction to conduct a protein-interaction-network-based pathway analysis (PINBPA) on two large genetic MS studies comprising a total of 15,317 cases and 29,529 controls. The distribution of nominally significant loci at the gene level matched the patterns of extended linkage disequilibrium in regions of interest. We found that products of genome-wide significantly associated genes are more likely to interact physically and belong to the same or related pathways. We next searched for subnetworks (modules) of genes (and their encoded proteins) enriched with nominally associated loci within each study and identified those modules in common between the two studies. We demonstrate that these modules are more likely to contain genes with bona fide susceptibility variants and, in addition, identify several high-confidence candidates (including BCL10, CD48, REL, TRAF3, and TEC). PINBPA is a powerful approach to gaining further insights into the biology of associated genes and to prioritizing candidates for subsequent genetic studies of complex traits.     

The Analysis of Sensory Data by Generalized Linear Model

The use of generalized linear model theory in formulating and testing hypotheses which may arise in the analysis of ordinal scale organoleptic data is illustrated.     

Insights into Colon Cancer Etiology via a Regularized Approach to Gene Set Analysis of GWAS Data

Genome-wide association studies (GWAS) have successfully identified susceptibility loci from marginal association analysis of SNPs. Valuable insight into genetic variation underlying complex diseases will likely be gained by considering functionally related sets of genes simultaneously. One approach is to further develop gene set enrichment analysis methods, which are initiated in gene expression studies, to account for the distinctive features of GWAS data. These features include the large number of SNPs per gene, the modest and sparse SNP associations, and the additional information provided by linkage disequilibrium (LD) patterns within genes. We propose a “gene set ridge regression in association studies (GRASS)” algorithm. GRASS summarizes the genetic structure for each gene as eigenSNPs and uses a novel form of regularized regression technique, termed group ridge regression, to select representative eigenSNPs for each gene and assess their joint association with disease risk. Compared with existing methods, the proposed algorithm greatly reduces the high dimensionality of GWAS data while still accounting for multiple hits and/or LD in the same gene. We show by simulation that this algorithm performs well in situations in which there are a large number of predictors compared to sample size. We applied the GRASS algorithm to a genome-wide association study of colon cancer and identified nicotinate and nicotinamide metabolism and transforming growth factor beta signaling as the top two significantly enriched pathways. Elucidating the role of variation in these pathways may enhance our understanding of colon cancer etiology.     

Mechanistic Phenotypes: An Aggregative Phenotyping Strategy to Identify Disease Mechanisms Using GWAS Data

A single mutation can alter cellular and global homeostatic mechanisms and give rise to multiple clinical diseases. We hypothesized that these disease mechanisms could be identified using low minor allele frequency (MAF<0.1) non-synonymous SNPs (nsSNPs) associated with “mechanistic phenotypes”, comprised of collections of related diagnoses. We studied two mechanistic phenotypes: (1) thrombosis, evaluated in a population of 1,655 African Americans; and (2) four groupings of cancer diagnoses, evaluated in 3,009 white European Americans. We tested associations between nsSNPs represented on GWAS platforms and mechanistic phenotypes ascertained from electronic medical records (EMRs), and sought enrichment in functional ontologies across the top-ranked associations. We used a two-step analytic approach whereby nsSNPs were first sorted by the strength of their association with a phenotype. We tested associations using two reverse genetic models and standard additive and recessive models. In the second step, we employed a hypothesis-free ontological enrichment analysis using the sorted nsSNPs to identify functional mechanisms underlying the diagnoses comprising the mechanistic phenotypes. The thrombosis phenotype was solely associated with ontologies related to blood coagulation (Fisher''s p = 0.0001, FDR p = 0.03), driven by the F5, P2RY12 and F2RL2 genes. For the cancer phenotypes, the reverse genetics models were enriched in DNA repair functions (p = 2×10−5, FDR p = 0.03) (POLG/FANCI, SLX4/FANCP, XRCC1, BRCA1, FANCA, CHD1L) while the additive model showed enrichment related to chromatid segregation (p = 4×10−6, FDR p = 0.005) (KIF25, PINX1). We were able to replicate nsSNP associations for POLG/FANCI, BRCA1, FANCA and CHD1L in independent data sets. Mechanism-oriented phenotyping using collections of EMR-derived diagnoses can elucidate fundamental disease mechanisms.     

Development of GMDR-GPU for Gene-Gene Interaction Analysis and Its Application to WTCCC GWAS Data for Type 2 Diabetes

Although genome-wide association studies (GWAS) have identified a significant number of single-nucleotide polymorphisms (SNPs) associated with many complex human traits, the susceptibility loci identified so far can explain only a small fraction of the genetic risk. Among other possible explanations, the lack of a comprehensive examination of gene–gene interaction (G×G) is often considered a source of the missing heritability. Previously, we reported a model-free Generalized Multifactor Dimensionality Reduction (GMDR) approach for detecting G×G in both dichotomous and quantitative phenotypes. However, the computational burden and less efficient implementation of the original programs make them impossible to use for GWAS. In this study, we developed a graphics processing unit (GPU)-based GMDR program (named GWAS-GPU), which is able not only to analyze GWAS data but also to run much faster than the earlier version of the GMDR program. As a demonstration of the program, we used the GMDR-GPU software to analyze a publicly available GWAS dataset on type 2 diabetes (T2D) from the Wellcome Trust Case Control Consortium. Through an exhaustive search of pair-wise interactions and a selected search of three- to five-way interactions conditioned on significant pair-wise results, we identified 24 core SNPs in six genes (FTO: rs9939973, rs9940128, rs9922047, rs1121980, rs9939609, rs9930506; TSPAN8: rs1495377; TCF7L2: rs4074720, rs7901695, rs4506565, rs4132670, rs10787472, rs11196205, rs10885409, rs11196208; L3MBTL3: rs10485400, rs4897366; CELF4: rs2852373, rs608489; RUNX1: rs445984, rs1040328, rs990074, rs2223046, rs2834970) that appear to be important for T2D. Of these core SNPs, 11 in FTO, TSPAN8, and TCF7L2 have been reported to be associated with T2D, obesity, or both, providing an independent replication of previously reported SNPs. Importantly, we identified three new susceptibility genes; i.e., L3MBTL3, CELF4, and RUNX1, for T2D, a finding that warrants further investigation with independent samples.     

Integrating GWAS and Expression Data for Functional Characterization of Disease-Associated SNPs: An Application to Follicular Lymphoma

Development of post-GWAS (genome-wide association study) methods are greatly needed for characterizing the function of trait-associated SNPs. Strategies integrating various biological data sets with GWAS results will provide insights into the mechanistic role of associated SNPs. Here, we present a method that integrates RNA sequencing (RNA-seq) and allele-specific expression data with GWAS data to further characterize SNPs associated with follicular lymphoma (FL). We investigated the influence on gene expression of three established FL-associated loci—rs10484561, rs2647012, and rs6457327—by measuring their correlation with human-leukocyte-antigen (HLA) expression levels obtained from publicly available RNA-seq expression data sets from lymphoblastoid cell lines. Our results suggest that SNPs linked to the protective variant rs2647012 exert their effect by a cis-regulatory mechanism involving modulation of HLA-DQB1 expression. In contrast, no effect on HLA expression was observed for the colocalized risk variant rs10484561. The application of integrative methods, such as those presented here, to other post-GWAS investigations will help identify causal disease variants and enhance our understanding of biological disease mechanisms.     

GWAS 2011: Opportunities and challenges

The wide adoption of genome-wide association study (GWAS) has dramatically changed the landscape of the genetic studies of human diseases.Banking on the study design that employs large and multiple-independent samples,linkage disequilibrium (LD) -based systematic genome-wide interrogation,and vigorous statistical standard for declaring genetic association,GWAS has greatly advanced the genetic studies of human disease by successfully identifying over 4 thousands of genetic susceptibility SNPs or loci for 210 diseases/traits in human.Besides its unrivaled scientific achievements,GWAS has also transformed the communities of human genetics by stimulating unprecedented broad and large-scale collaboration and data sharing across different groups and countries,which has enabled some mega genetic studies where dozens or hundreds of thousands of samples were investigated through mete-analysis,providing enormous statistical power for discovering genetic variants,even the ones with very moderate effect on disease phenotype or physiological trait.While the success of GWAS has been widely recognized,there are many concerns on GWAS that are being passionately debated.In particular,there ate heated discussions on why only a limited proportion,quiet often a small proportion,of genetic heritability can be explained by the GWAS findings,and whether and how the GWAS findings have really advanced the biological investigation and understanding of disease mechanisms.Concerning these shortcomings of GWAS,many have genuinely questioned the perspective of translating the GWAS findings into clinical practice.To provide a platform to discuss the progresses and shortcomings of GWAS,a nature conference,titled “GWAS 2011:Opportunities and Challenges”,was organized through the collaboration between the Nature Genetics and Anhui Medical University,which was held in Hefei,China on May 19-21,2011.     

META-GSA: Combining Findings from Gene-Set Analyses across Several Genome-Wide Association Studies

Introduction

Gene-set analysis (GSA) methods are used as complementary approaches to genome-wide association studies (GWASs). The single marker association estimates of a predefined set of genes are either contrasted with those of all remaining genes or with a null non-associated background. To pool the p-values from several GSAs, it is important to take into account the concordance of the observed patterns resulting from single marker association point estimates across any given gene set. Here we propose an enhanced version of Fisher’s inverse χ²-method META-GSA, however weighting each study to account for imperfect correlation between association patterns.

Simulation and Power

We investigated the performance of META-GSA by simulating GWASs with 500 cases and 500 controls at 100 diallelic markers in 20 different scenarios, simulating different relative risks between 1 and 1.5 in gene sets of 10 genes. Wilcoxon’s rank sum test was applied as GSA for each study. We found that META-GSA has greater power to discover truly associated gene sets than simple pooling of the p-values, by e.g. 59% versus 37%, when the true relative risk for 5 of 10 genes was assume to be 1.5. Under the null hypothesis of no difference in the true association pattern between the gene set of interest and the set of remaining genes, the results of both approaches are almost uncorrelated. We recommend not relying on p-values alone when combining the results of independent GSAs.

Application

We applied META-GSA to pool the results of four case-control GWASs of lung cancer risk (Central European Study and Toronto/Lunenfeld-Tanenbaum Research Institute Study; German Lung Cancer Study and MD Anderson Cancer Center Study), which had already been analyzed separately with four different GSA methods (EASE; SLAT, mSUMSTAT and GenGen). This application revealed the pathway GO0015291 “transmembrane transporter activity” as significantly enriched with associated genes (GSA-method: EASE, p = 0.0315 corrected for multiple testing). Similar results were found for GO0015464 “acetylcholine receptor activity” but only when not corrected for multiple testing (all GSA-methods applied; p≈0.02).     

Enrichment Map: A Network-Based Method for Gene-Set Enrichment Visualization and Interpretation

Background

Gene-set enrichment analysis is a useful technique to help functionally characterize large gene lists, such as the results of gene expression experiments. This technique finds functionally coherent gene-sets, such as pathways, that are statistically over-represented in a given gene list. Ideally, the number of resulting sets is smaller than the number of genes in the list, thus simplifying interpretation. However, the increasing number and redundancy of gene-sets used by many current enrichment analysis software works against this ideal.

Principal Findings

To overcome gene-set redundancy and help in the interpretation of large gene lists, we developed “Enrichment Map”, a network-based visualization method for gene-set enrichment results. Gene-sets are organized in a network, where each set is a node and edges represent gene overlap between sets. Automated network layout groups related gene-sets into network clusters, enabling the user to quickly identify the major enriched functional themes and more easily interpret the enrichment results.

Conclusions

Enrichment Map is a significant advance in the interpretation of enrichment analysis. Any research project that generates a list of genes can take advantage of this visualization framework. Enrichment Map is implemented as a freely available and user friendly plug-in for the Cytoscape network visualization software (http://baderlab.org/Software/EnrichmentMap/).     

Large Scale Analysis of Phenotype-Pathway Relationships Based on GWAS Results

The widely used pathway-based approach for interpreting Genome Wide Association Studies (GWAS), assumes that since function is executed through the interactions of multiple genes, different perturbations of the same pathway would result in a similar phenotype. This assumption, however, was not systemically assessed on a large scale. To determine whether SNPs associated with a given complex phenotype affect the same pathways more than expected by chance, we analyzed 368 phenotypes that were studied in >5000 GWAS. We found 216 significant phenotype-pathway associations between 70 of the phenotypes we analyzed and known pathways. We also report 391 strong phenotype-phenotype associations between phenotypes that are affected by the same pathways. While some of these associations confirm previously reported connections, others are new and could shed light on the molecular basis of these diseases. Our findings confirm that phenotype-associated SNPs cluster into pathways much more than expected by chance. However, this is true for <20% (70/368) of the phenotypes. Different types of phenotypes show markedly different tendencies: Virtually all autoimmune phenotypes show strong clustering of SNPs into pathways, while most cancers and metabolic conditions, and all electrophysiological phenotypes, could not be significantly associated with any pathway despite being significantly associated with a large number of SNPs. While this may be due to missing data, it may also suggest that these phenotypes could result only from perturbations of specific genes and not from other perturbations of the same pathway. Further analysis of pathway-associated versus gene-associated phenotypes is, therefore, needed in order to understand disease etiology and in order to promote better drug target selection.     

Connecting the Dots: Potential of Data Integration to Identify Regulatory SNPs in Late-Onset Alzheimer's Disease GWAS Findings

Late-onset Alzheimer''s disease (LOAD) is a multifactorial disorder with over twenty loci associated with disease risk. Given the number of genome-wide significant variants that fall outside of coding regions, it is possible that some of these variants alter some function of gene expression rather than tagging coding variants that alter protein structure and/or function. RegulomeDB is a database that annotates regulatory functions of genetic variants. In this study, we utilized RegulomeDB to investigate potential regulatory functions of lead single nucleotide polymorphisms (SNPs) identified in five genome-wide association studies (GWAS) of risk and age-at onset (AAO) of LOAD, as well as SNPs in LD (r²≥0.80) with the lead GWAS SNPs. Of a total 614 SNPs examined, 394 returned RegulomeDB scores of 1–6. Of those 394 variants, 34 showed strong evidence of regulatory function (RegulomeDB score <3), and only 3 of them were genome-wide significant SNPs (ZCWPW1/rs1476679, CLU/rs1532278 and ABCA7/rs3764650). This study further supports the assumption that some of the non-coding GWAS SNPs are true associations rather than tagged associations and demonstrates the application of RegulomeDB to GWAS data.     

GENOVA: gene overlap analysis of GWAS results

In many published genome-wide association studies (GWAS), the top few strongly associated variants are often located in or near known genes. This observation raises the more general hypothesis that variants nominally associated with a phenotype are more likely to overlap genes than those not associated with a phenotype. We developed a simple approach - named GENe OVerlap Analysis (GENOVA) - to formally test this hypothesis. This approach includes two steps. First, we define largely independent groups of highly correlated SNPs (or "clumps") and classify each clump as intersecting a gene or not. Second, we determine how strongly associated each clump is with the phenotype and use logistic regression to formally test the hypothesis that clumps associated with the phenotype are more likely to intersect genes. Simulations suggest that the power of GENOVA is affected by at least three factors: GWAS sample size, the gene boundaries used to define gene-intersecting clumps and the P-value threshold used to define phenotype-associated clumps. We applied GENOVA to results from three recent GWAS meta-analyses of height, body mass index (BMI) and waist-hip ratio (WHR) conducted by the GIANT consortium. SNPs associated with variation in height were 1.44-fold more likely to be in or near genes than SNPs not associated with height (P = 5 x 10?2?). A weaker association was observed for BMI (1.09-fold, P = 0.008) and WHR (1.09-fold, P = 0.014). GENOVA is implemented in C++ and is freely available at https://genepi.qimr.edu.au/staff/manuelF/genova/main.html.