96
R. E. Graff et al.
5.2.4 Transcriptome-Wide Association Studies
Among the more recent methodological developments in genetic association studies
is the transcriptome-wide association study (TWAS) (Gamazon et al. 2015; Gusev
et al. 2016). Without relying on directly measured expression levels, TWAS aim to
identify genes associated with complex traits. By using an external reference set of
individuals with genetic and transcriptomic data, one can impute gene expression
levels in the target study population and evaluate associations with the outcome.
Extensions of this approach allow for implementation with summary statistics rather
than individual-level data, making TWAS an increasingly popular study design
(Barbeira et al. 2018). Furthermore, because associations at the gene expression
level often have clearer functional interpretations than associations with individual
risk variants, TWAS have the potential to offer insights distinct from those offered
by GWAS. Testing for associations with genes rather than SNPs also reduces
the multiple testing burden, thereby improving statistical power for discovery.
TWAS are, however, limited by the comprehensiveness of gene expression reference
panels both across different tissues and for populations of non-European ancestry.
Furthermore, although the genetic architecture of gene expression allows for
reasonable imputation accuracy, gene expression can also be influenced by nongenetic, external factors.
5.2.5 Replication and Meta-analysis
Findings from a single genetic association study are not generally sufficient to instill
confidence in results. Rather, results should be validated in independent samples and
combined with other studies to bolster sample size.
5.2.5.1 Replication
Early genetic association studies frequently yielded results that failed to reproduce
in independent samples (Hirschhorn et al. 2002; Ioannidis 2006; Ioannidis et al.
2001; Lohmueller et al. 2003). Why the surfeit of false positives? Historically,
studies of candidate markers or genes often had small sample sizes, inappropriate
thresholds for statistical significance, and/or low prior probabilities of true associations (Chanock et al. 2007; Hirschhorn and Altshuler 2002; Ioannidis 2005; Manolio
et al. 2008; Mutsuddi et al. 2006; Wacholder et al. 2004). Even now, investigators are
conscious of “winner’s curse,” whereby the effect estimates from initial discovery
studies are consistently biased upward (Lohmueller et al. 2003; Goring et al.
2001; Huang et al. 2018). They are generally more attentive to winner’s curse for
genetic association studies than for other epidemiological investigations because
the former most often test a large number of exposures. The gold standard for
substantiating results from genetic association studies has thus become replication
in independent samples. Replication has become important (and essentially required
for publication) to externally validate the credibility of genetic associations.
R. E. Graff et al.
5.2.4 Transcriptome-Wide Association Studies
Among the more recent methodological developments in genetic association studies
is the transcriptome-wide association study (TWAS) (Gamazon et al. 2015; Gusev
et al. 2016). Without relying on directly measured expression levels, TWAS aim to
identify genes associated with complex traits. By using an external reference set of
individuals with genetic and transcriptomic data, one can impute gene expression
levels in the target study population and evaluate associations with the outcome.
Extensions of this approach allow for implementation with summary statistics rather
than individual-level data, making TWAS an increasingly popular study design
(Barbeira et al. 2018). Furthermore, because associations at the gene expression
level often have clearer functional interpretations than associations with individual
risk variants, TWAS have the potential to offer insights distinct from those offered
by GWAS. Testing for associations with genes rather than SNPs also reduces
the multiple testing burden, thereby improving statistical power for discovery.
TWAS are, however, limited by the comprehensiveness of gene expression reference
panels both across different tissues and for populations of non-European ancestry.
Furthermore, although the genetic architecture of gene expression allows for
reasonable imputation accuracy, gene expression can also be influenced by nongenetic, external factors.
5.2.5 Replication and Meta-analysis
Findings from a single genetic association study are not generally sufficient to instill
confidence in results. Rather, results should be validated in independent samples and
combined with other studies to bolster sample size.
5.2.5.1 Replication
Early genetic association studies frequently yielded results that failed to reproduce
in independent samples (Hirschhorn et al. 2002; Ioannidis 2006; Ioannidis et al.
2001; Lohmueller et al. 2003). Why the surfeit of false positives? Historically,
studies of candidate markers or genes often had small sample sizes, inappropriate
thresholds for statistical significance, and/or low prior probabilities of true associations (Chanock et al. 2007; Hirschhorn and Altshuler 2002; Ioannidis 2005; Manolio
et al. 2008; Mutsuddi et al. 2006; Wacholder et al. 2004). Even now, investigators are
conscious of “winner’s curse,” whereby the effect estimates from initial discovery
studies are consistently biased upward (Lohmueller et al. 2003; Goring et al.
2001; Huang et al. 2018). They are generally more attentive to winner’s curse for
genetic association studies than for other epidemiological investigations because
the former most often test a large number of exposures. The gold standard for
substantiating results from genetic association studies has thus become replication
in independent samples. Replication has become important (and essentially required
for publication) to externally validate the credibility of genetic associations.