102
R. E. Graff et al.
the variants be previously ascertained and included on genome-wide genotyping
arrays. Originally, such technologies were most often utilized for (and successful
at) identifying genetic contributors to many Mendelian disorders. More recently,
the technologies have been leveraged to assay the exome as a popular approach
for evaluating associations between rare variants and complex phenotypes. Exome
sequencing selects the entire set of human exons as the sequencing target (Gnirke
et al. 2009; Hodges et al. 2007). In contrast, exome arrays concentrate on a fixed
set of variants. Regardless of the platform used to evaluate the exome, the variants
assessed have functional implications that are relatively easy to derive. Still, due to
reduced penetrance, sample sizes required for detecting associations with complex
traits are generally larger than those required for the evaluation of Mendelian
disorders. As exome sequencing costs have come down, however, it has become
increasingly feasible to conduct well-powered studies. It is just important that they
increase sample sizes in proportion to the rarity of causal variants.
5.5
Data Analysis
Upon completing the measurement of both genotype and phenotype, one must
consider the appropriate methods for the analysis of the data. In addition to
thinking through the statistical methods that should be applied, one must also assess
the data for their quality, consider covariates that should be accounted for, and
potentially incorporate information from external sources. Below we identify some
key considerations for analyzing genetic association studies.
5.5.1 Quality Control
Before analyzing the data from any study of genetic association, it is imperative
that the genotyping be subject to a number of quality control checks. Samples that
come from various sources may be processed in different ways or measured at
different times, which can result in systematic differences across batches. One must
also evaluate the proportion of samples that are successfully genotyped and test for
Hardy–Weinberg equilibrium. Issues with these metrics could indicate genotyping
problems that affect all of the SNPs in the sample. As such, one should remove
SNPs that fail predefined quality standards from further consideration. Using SNP
genotypes and external LD information on the underlying structure of a genetic
region (e.g., from the TOPMed imputation reference panel), one can impute some
of the untyped variants and variants that fail quality control. Doing so allows for a
more thorough and powerful evaluation of potential associations across the genome
(Huang et al. 2009; Marchini and Howie 2008; Marchini et al. 2007).
Note that sequencing (as opposed to genotyping) also requires appreciable
quality control efforts, but their description is beyond the scope of this chapter.
R. E. Graff et al.
the variants be previously ascertained and included on genome-wide genotyping
arrays. Originally, such technologies were most often utilized for (and successful
at) identifying genetic contributors to many Mendelian disorders. More recently,
the technologies have been leveraged to assay the exome as a popular approach
for evaluating associations between rare variants and complex phenotypes. Exome
sequencing selects the entire set of human exons as the sequencing target (Gnirke
et al. 2009; Hodges et al. 2007). In contrast, exome arrays concentrate on a fixed
set of variants. Regardless of the platform used to evaluate the exome, the variants
assessed have functional implications that are relatively easy to derive. Still, due to
reduced penetrance, sample sizes required for detecting associations with complex
traits are generally larger than those required for the evaluation of Mendelian
disorders. As exome sequencing costs have come down, however, it has become
increasingly feasible to conduct well-powered studies. It is just important that they
increase sample sizes in proportion to the rarity of causal variants.
5.5
Data Analysis
Upon completing the measurement of both genotype and phenotype, one must
consider the appropriate methods for the analysis of the data. In addition to
thinking through the statistical methods that should be applied, one must also assess
the data for their quality, consider covariates that should be accounted for, and
potentially incorporate information from external sources. Below we identify some
key considerations for analyzing genetic association studies.
5.5.1 Quality Control
Before analyzing the data from any study of genetic association, it is imperative
that the genotyping be subject to a number of quality control checks. Samples that
come from various sources may be processed in different ways or measured at
different times, which can result in systematic differences across batches. One must
also evaluate the proportion of samples that are successfully genotyped and test for
Hardy–Weinberg equilibrium. Issues with these metrics could indicate genotyping
problems that affect all of the SNPs in the sample. As such, one should remove
SNPs that fail predefined quality standards from further consideration. Using SNP
genotypes and external LD information on the underlying structure of a genetic
region (e.g., from the TOPMed imputation reference panel), one can impute some
of the untyped variants and variants that fail quality control. Doing so allows for a
more thorough and powerful evaluation of potential associations across the genome
(Huang et al. 2009; Marchini and Howie 2008; Marchini et al. 2007).
Note that sequencing (as opposed to genotyping) also requires appreciable
quality control efforts, but their description is beyond the scope of this chapter.