Assessing the impact of differential genotyping errors on rare variant tests of association

Establishing analytical validity of BeadChip array genotype data by comparison to whole-genome sequence and standard benchmark datasets

BACKGROUND

Clinical use of genotype data requires high positive predictive value (PPV) and thorough understanding of the genotyping platform characteristics. BeadChip arrays, such as the Global Screening Array (GSA), potentially offer a high-throughput, low-cost clinical screen for known variants. We hypothesize that quality assessment and comparison to whole-genome sequence and benchmark data establish the analytical validity of GSA genotyping.

METHODS

To test this hypothesis, we selected 263 samples from Coriell, generated GSA genotypes in triplicate, generated whole genome sequence (rWGS) genotypes, assessed the quality of each set of genotypes, and compared each set of genotypes to each other and to the 1000 Genomes Phase 3 (1KG) genotypes, a performance benchmark. For 59 genes (MAP59), we also performed theoretical and empirical evaluation of variants deemed medically actionable predispositions.

RESULTS

Quality analyses detected sample contamination and increased assay failure along the chip margins. Comparison to benchmark data demonstrated that > 82% of the GSA assays had a PPV of 1. GSA assays targeting transitions, genomic regions of high complexity, and common variants performed better than those targeting transversions, regions of low complexity, and rare variants. Comparison of GSA data to rWGS and 1KG data showed > 99% performance across all measured parameters. Consistent with predictions from prior studies, the GSA detection of variation within the MAP59 genes was 3/261.

CONCLUSION

We establish the analytical validity of GSA assays using quality analytics and comparison to benchmark and rWGS data. GSA assays meet the standards of a clinical screen although assays interrogating rare variants, transversions, and variants within low-complexity regions require careful evaluation.

Comparative Analysis of SNP Discovery and Genotyping in Fagus sylvatica L. and Quercus robur L. Using RADseq, GBS, and ddRAD Methods

Next-generation sequencing of reduced representation genomic libraries (RRL) is capable of providing large numbers of genetic markers for population genetic studies at relatively low costs. However, one major concern of these types of markers is the precision of genotyping, which is related to the common problem of missing data, which appears to be particularly important in association and genomic selection studies. We evaluated three RRL approaches (GBS, RADseq, ddRAD) and different SNP identification methods (de novo or based on a reference genome) to find the best solutions for future population genomics studies in two economically and ecologically important broadleaved tree species, namely F. sylvatica and Q. robur. We found that the use of ddRAD method coupled with SNP calling based on reference genomes provided the largest numbers of markers (28 k and 36 k for beech and oak, respectively), given standard filtering criteria. Using technical replicates of samples, we demonstrated that more than 80% of SNP loci should be considered as reliable markers in GBS and ddRAD, but not in RADseq data. According to the reference genomes’ annotations, more than 30% of the identified ddRAD loci appeared to be related to genes. Our findings provide a solid support for using ddRAD-based SNPs for future population genomics studies in beech and oak.

Impact of genotyping errors on statistical power of association tests in genomic analyses: A case study

A key step in genomic studies is to assess high throughput measurements across millions of markers for each participant's DNA, either using microarrays or sequencing techniques. Accurate genotype calling is essential for downstream statistical analysis of genotype-phenotype associations, and next generation sequencing (NGS) has recently become a more common approach in genomic studies. How the accuracy of variant calling in NGS-based studies affects downstream association analysis has not, however, been studied using empirical data in which both microarrays and NGS were available. In this article, we investigate the impact of variant calling errors on the statistical power to identify associations between single nucleotides and disease, and on associations between multiple rare variants and disease. Both differential and nondifferential genotyping errors are considered. Our results show that the power of burden tests for rare variants is strongly influenced by the specificity in variant calling, but is rather robust with regard to sensitivity. By using the variant calling accuracies estimated from a substudy of a Cooperative Studies Program project conducted by the Department of Veterans Affairs, we show that the power of association tests is mostly retained with commonly adopted variant calling pipelines. An R package, GWAS.PC, is provided to accommodate power analysis that takes account of genotyping errors (http://zhaocenter.org/software/).

A POWERFUL METHOD FOR INCLUDING GENOTYPE UNCERTAINTY IN TESTS OF HARDY-WEINBERG EQUILIBRIUM

The use of posterior probabilities to summarize genotype uncertainty is pervasive across genotype, sequencing and imputation platforms. Prior work in many contexts has shown the utility of incorporating genotype uncertainty (posterior probabilities) in downstream statistical tests. Typical approaches to incorporating genotype uncertainty when testing Hardy-Weinberg equilibrium tend to lack calibration in the type I error rate, especially as genotype uncertainty increases. We propose a new approach in the spirit of genomic control that properly calibrates the type I error rate, while yielding improved power to detect deviations from Hardy-Weinberg Equilibrium. We demonstrate the improved performance of our method on both simulated and real genotypes.

Phenotypically Enriched Genotypic Imputation in Genetic Association Tests

BACKGROUND

In longitudinal epidemiological studies there may be individuals with rich phenotype data who die or are lost to follow-up before providing DNA for genetic studies. Often, the genotypic and phenotypic data of the relatives are available. Two strategies for analyzing the incomplete data are to exclude ungenotyped subjects from analysis (the complete-case method, CC) and to include phenotyped but ungenotyped individuals in analysis by using relatives' genotypes for genotype imputation (GI). In both strategies, the information in the phenotypic data was not used to handle the missing-genotype problem.

METHODS

We propose a phenotypically enriched genotypic imputation (PEGI) method that uses the EM (expectation-maximization)-based maximum likelihood method to incorporate observed phenotypes into genotype imputation.

RESULTS

Our simulations with genotypes missing completely at random show that, for a single-nucleotide polymorphism (SNP) with moderate to strong effect on a phenotype, PEGI improves power more than GI without excess type I errors. Using the Framingham Heart Study data set, we compare the ability of the PEGI, GI, and CC to detect the associations between 5 SNPs and age at natural menopause.

CONCLUSION

The PEGI method may improve power to detect an association over both CC and GI under many circumstances.

The impact of genotype calling errors on family-based studies

Family-based sequencing studies have unique advantages in enriching rare variants, controlling population stratification, and improving genotype calling. Standard genotype calling algorithms are less likely to call rare variants correctly, often mistakenly calling heterozygotes as reference homozygotes. The consequences of such non-random errors on association tests for rare variants are unclear, particularly in transmission-based tests. In this study, we investigated the impact of genotyping errors on rare variant association tests of family-based sequence data. We performed a comprehensive analysis to study how genotype calling errors affect type I error and statistical power of transmission-based association tests using a variety of realistic parameters in family-based sequencing studies. In simulation studies, we found that biased genotype calling errors yielded not only an inflation of type I error but also a power loss of association tests. We further confirmed our observation using exome sequence data from an autism project. We concluded that non-symmetric genotype calling errors need careful consideration in the analysis of family-based sequence data and we provided practical guidance on ameliorating the test bias.

Testing Rare-Variant Association without Calling Genotypes Allows for Systematic Differences in Sequencing between Cases and Controls

Next-generation sequencing of DNA provides an unprecedented opportunity to discover rare genetic variants associated with complex diseases and traits. However, the common practice of first calling underlying genotypes and then treating the called values as known is prone to false positive findings, especially when genotyping errors are systematically different between cases and controls. This happens whenever cases and controls are sequenced at different depths, on different platforms, or in different batches. In this article, we provide a likelihood-based approach to testing rare variant associations that directly models sequencing reads without calling genotypes. We consider the (weighted) burden test statistic, which is the (weighted) sum of the score statistic for assessing effects of individual variants on the trait of interest. Because variant locations are unknown, we develop a simple, computationally efficient screening algorithm to estimate the loci that are variants. Because our burden statistic may not have mean zero after screening, we develop a novel bootstrap procedure for assessing the significance of the burden statistic. We demonstrate through extensive simulation studies that the proposed tests are robust to a wide range of differential sequencing qualities between cases and controls, and are at least as powerful as the standard genotype calling approach when the latter controls type I error. An application to the UK10K data reveals novel rare variants in gene BTBD18 associated with childhood onset obesity. The relevant software is freely available.

In next-generation sequencing studies, there are typically systematic differences in sequencing qualities (e.g., depth) between cases and controls, because the entire studies are rarely sequenced in exactly the same way. It has long been appreciated that, in the presence of such differences, the standard genotype calling approach to detecting rare variant associations generally leads to excessive false positive findings. To deal with this, the current “state of the art” is to impose stringent quality control procedures that much of the data is eliminated. We present a method that allows analyzing data with a wide range of differential sequencing qualities between cases and controls. Our method is more powerful than the current practice and can accelerate the search for disease-causing mutations.

Whole-genome sequence data uncover loss of genetic diversity due to selection

BACKGROUND

Whole-genome sequence (WGS) data give access to more complete structural genetic information of individuals, including rare variants, not fully covered by single nucleotide polymorphism chips. We used WGS to investigate the amount of genetic diversity remaining after selection using optimal contribution (OC), considering different methods to estimate the relationships used in OC. OC was applied to minimise average relatedness of the selection candidates and thus miminise the loss of genetic diversity in a conservation strategy, e.g. for establishment of gene bank collections. Furthermore, OC was used to maximise average genetic merit of the selection candidates at a given level of relatedness, similar to a genetic improvement strategy. In this study, we used data from 277 bulls from the 1000 bull genomes project. We measured genetic diversity as the number of variants still segregating after selection using WGS data, and compared strategies that targeted conservation of rare (minor allele frequency <5 %) versus common variants.

RESULTS

When OC without restriction on the number of selected individuals was applied, loss of variants was minimal and most individuals were selected, which is often unfeasible in practice. When 20 individuals were selected, the number of segregating rare variants was reduced by 29 % for the conservation strategy, and by 34 % for the genetic improvement strategy. The overall number of segregating variants was reduced by 30 % when OC was restricted to selecting five individuals, for both conservation and genetic improvement strategies. For common variants, this loss was about 15 %, while it was much higher, 72 %, for rare variants. Fewer rare variants were conserved with the genetic improvement strategy compared to the conservation strategy.

CONCLUSIONS

The use of WGS for genetic diversity quantification revealed that selection results in considerable losses of genetic diversity for rare variants. Using WGS instead of SNP chip data to estimate relationships slightly reduced the loss of rare variants, while using 50 K SNP chip data was sufficient to conserve common variants. The loss of rare variants could be mitigated by a few percent (up to 8 %) depending on which method is chosen to estimate relationships from WGS data.

Value of Mendelian laws of segregation in families: data quality control, imputation, and beyond

When analyzing family data, we dream of perfectly informative data, even whole-genome sequences (WGSs) for all family members. Reality intervenes, and we find that next-generation sequencing (NGS) data have errors and are often too expensive or impossible to collect on everyone. The Genetic Analysis Workshop 18 working groups on quality control and dropping WGSs through families using a genome-wide association framework focused on finding, correcting, and using errors within the available sequence and family data, developing methods to infer and analyze missing sequence data among relatives, and testing for linkage and association with simulated blood pressure. We found that single-nucleotide polymorphisms, NGS data, and imputed data are generally concordant but that errors are particularly likely at rare variants, for homozygous genotypes, within regions with repeated sequences or structural variants, and within sequence data imputed from unrelated individuals. Admixture complicated identification of cryptic relatedness, but information from Mendelian transmission improved error detection and provided an estimate of the de novo mutation rate. Computationally, fast rule-based imputation was accurate but could not cover as many loci or subjects as more computationally demanding probability-based methods. Incorporating population-level data into pedigree-based imputation methods improved results. Observed data outperformed imputed data in association testing, but imputed data were also useful. We discuss the strengths and weaknesses of existing methods and suggest possible future directions, such as improving communication between data collectors and data analysts, establishing thresholds for and improving imputation quality, and incorporating error into imputation and analytical models.

Evaluating the concordance between sequencing, imputation and microarray genotype calls in the GAW18 data

Genotype errors are well known to increase type I errors and/or decrease power in related tests of genotype-phenotype association, depending on whether the genotype error mechanism is associated with the phenotype. These relationships hold for both single and multimarker tests of genotype-phenotype association. To assess the potential for genotype errors in Genetic Analysis Workshop 18 (GAW18) data, where no gold standard genotype calls are available, we explored concordance rates between sequencing, imputation, and microarray genotype calls. Our analysis shows that missing data rates for sequenced individuals are high and that there is a modest amount of called genotype discordance between the 2 platforms, with discordance most common for lower minor allele frequency (MAF) single-nucleotide polymorphisms (SNPs). Some evidence for discordance rates that were different between phenotypes was observed, and we identified a number of cases where different technologies identified different bases at the variant site. Type I errors and power loss is possible as a result of missing genotypes and errors in called genotypes in downstream analysis of GAW18 data.

Effective filtering strategies to improve data quality from population-based whole exome sequencing studies

Background

Genotypes generated in next generation sequencing studies contain errors which can significantly impact the power to detect signals in common and rare variant association tests. These genotyping errors are not explicitly filtered by the standard GATK Variant Quality Score Recalibration (VQSR) tool and thus remain a source of errors in whole exome sequencing (WES) projects that follow GATK’s recommended best practices. Therefore, additional data filtering methods are required to effectively remove these errors before performing association analyses with complex phenotypes. Here we empirically derive thresholds for genotype and variant filters that, when used in conjunction with the VQSR tool, achieve higher data quality than when using VQSR alone.

Results

The detailed filtering strategies improve the concordance of sequenced genotypes with array genotypes from 99.33% to 99.77%; improve the percent of discordant genotypes removed from 10.5% to 69.5%; and improve the Ti/Tv ratio from 2.63 to 2.75. We also demonstrate that managing batch effects by separating samples based on different target capture and sequencing chemistry protocols results in a final data set containing 40.9% more high-quality variants. In addition, imputation is an important component of WES studies and is used to estimate common variant genotypes to generate additional markers for association analyses. As such, we demonstrate filtering methods for imputed data that improve genotype concordance from 79.3% to 99.8% while removing 99.5% of discordant genotypes.

Conclusions

The described filtering methods are advantageous for large population-based WES studies designed to identify common and rare variation associated with complex diseases. Compared to data processed through standard practices, these strategies result in substantially higher quality data for common and rare association analyses.

Evaluating the impact of genotype errors on rare variant tests of association

The new class of rare variant tests has usually been evaluated assuming perfect genotype information. In reality, rare variant genotypes may be incorrect, and so rare variant tests should be robust to imperfect data. Errors and uncertainty in SNP genotyping are already known to dramatically impact statistical power for single marker tests on common variants and, in some cases, inflate the type I error rate. Recent results show that uncertainty in genotype calls derived from sequencing reads are dependent on several factors, including read depth, calling algorithm, number of alleles present in the sample, and the frequency at which an allele segregates in the population. We have recently proposed a general framework for the evaluation and investigation of rare variant tests of association, classifying most rare variant tests into one of two broad categories (length or joint tests). We use this framework to relate factors affecting genotype uncertainty to the power and type I error rate of rare variant tests. We find that non-differential genotype errors (an error process that occurs independent of phenotype) decrease power, with larger decreases for extremely rare variants, and for the common homozygote to heterozygote error. Differential genotype errors (an error process that is associated with phenotype status), lead to inflated type I error rates which are more likely to occur at sites with more common homozygote to heterozygote errors than vice versa. Finally, our work suggests that certain rare variant tests and study designs may be more robust to the inclusion of genotype errors. Further work is needed to directly integrate genotype calling algorithm decisions, study costs and test statistic choices to provide comprehensive design and analysis advice which appropriately accounts for the impact of genotype errors.

Comparison of genotype clustering tools with rare variants

Background

Along with the improvement of high throughput sequencing technologies, the genetics community is showing marked interest for the rare variants/common diseases hypothesis. While sequencing can still be prohibitive for large studies, commercially available genotyping arrays targeting rare variants prove to be a reasonable alternative. A technical challenge of array based methods is the task of deriving genotype classes (homozygous or heterozygous) by clustering intensity data points. The performance of clustering tools for common polymorphisms is well established, while their performance when conducted with a large proportion of rare variants (where data points are sparse for genotypes containing the rare allele) is less known. We have compared the performance of four clustering tools (GenCall, GenoSNP, optiCall and zCall) for the genotyping of over 10,000 samples using the Illumina’s HumanExome BeadChip, which includes 247,870 variants, 90% of which have a minor allele frequency below 5% in a population of European ancestry. Different reference parameters for GenCall and different initial parameters for GenoSNP were tested. Genotyping accuracy was assessed using data from the 1000 Genomes Project as a gold standard, and agreement between tools was measured.

Results

Concordance of GenoSNP’s calls with the gold standard was below expectations and was increased by changing the tool’s initial parameters. While the four tools provided concordance with the gold standard above 99% for common alleles, some of them performed poorly for rare alleles. The reproducibility of genotype calls for each tool was assessed using experimental duplicates which provided concordance rates above 99%. The inter-tool agreement of genotype calls was high for approximately 95% of variants. Most tools yielded similar error rates (approximately 0.02), except for zCall which performed better with a 0.00164 mean error rate.

Conclusions

The GenoSNP clustering tool could not be run straight “out of the box” with the HumanExome BeadChip, as modification of hard coded parameters was necessary to achieve optimal performance. Overall, GenCall marginally outperformed the other tools for the HumanExome BeadChip. The use of experimental replicates provided a valuable quality control tool for genotyping projects with rare variants.

A geometric framework for evaluating rare variant tests of association

The wave of next-generation sequencing data has arrived. However, many questions still remain about how to best analyze sequence data, particularly the contribution of rare genetic variants to human disease. Numerous statistical methods have been proposed to aggregate association signals across multiple rare variant sites in an effort to increase statistical power; however, the precise relation between the tests is often not well understood. We present a geometric representation for rare variant data in which rare allele counts in case and control samples are treated as vectors in Euclidean space. The geometric framework facilitates a rigorous classification of existing rare variant tests into two broad categories: tests for a difference in the lengths of the case and control vectors, and joint tests for a difference in either the lengths or angles of the two vectors. We demonstrate that genetic architecture of a trait, including the number and frequency of risk alleles, directly relates to the behavior of the length and joint tests. Hence, the geometric framework allows prediction of which tests will perform best under different disease models. Furthermore, the structure of the geometric framework immediately suggests additional classes and types of rare variant tests. We consider two general classes of tests which show robustness to noncausal and protective variants. The geometric framework introduces a novel and unique method to assess current rare variant methodology and provides guidelines for both applied and theoretical researchers.