Plight of the GWAS

In an attempt to better understand human genetics, the genome-wide association study was developed, or the GWAS. The GWAS is a large-scale study which searches for particular gene variations associated with a specific condition. For example, a GWAS might be done to identify four possible alleles that contribute to increased risk for breast cancer.

The GWAS was expected to provide breakthroughs in genetic research by identifying significant numbers of genes to explain hereditary characteristics like height. Unfortunately, the results have been less than satisfactory, as Brendan Maher explains:

“GWAS, one of the most celebrated techniques of the past five years, promised to deliver many of the genes involved… And to some extent they have, identifying more than 400 genetic variants that contribute to a variety of traits and common diseases. But even when dozens of genes have been linked to a trait, both the individual and cumulative effects are disappointingly small and nowhere near enough to explain earlier estimates of heritability.”

Maher claims that height is generally believed to be 80-90% heritable. Despite this, the GWAS studies done on over 30,000 subjects were only able to predict about 5% of that height based on the genes they found.

This gap between GWAS results and alleged heritability isn’t only found in height. Only 2.5% of asthma risk can be explained by known genes, despite 25 studies done on over 150,000 individuals. GWAS results explain 6% of heritability for type 2 diabetes and just 1.5% for fasting glucose levels. This is called the “missing heritability” problem, and has yet to be conclusively resolved. The GWAS has failed to provide a genetic foundation for many important traits.

The Omnigenetic Model

Some geneticists argue the GWAS has inherent limitations that prevent scientists from predicting more of a characteristic’s heritability on an individual’s genetic makeup. They propose a situation where a characteristic is non-Mendelian to the tune of thousands of gene variants, each contributing a miniscule amount. The GWAS would be unable to distinguish such minute contributions from statistical noise, they argue.

This argument is, unfortunately, not supported by the scientific method — rather, it’s an explanation of why the scientific method fails in genetics. There is no currently available test for “very-low-penetrance variants,” and because these characteristics must (or so studies claim) be heritable, this is often the accepted scientific consensus, despite the inability to test the claim.

While later research might partially substantiate this so-called omnigenetic model, which was proposed by a group of Stanford geneticists, as of today, the theory is untestable. It’s wrong for scientists, or even members of the public, to point to the omnigenetic model as evidence of genetic traits beyond the scope of today’s science. One cannot claim athletic ability is 50% genetic and simultaneously cite the omnigenetic model. That claim is nothing but conjecture, ultimately doing little to help the actual science in genetics.

Significant non-genetic factors play into heritability, including epigenetic factors and the environment. There ought to be more questions as to the true genetic influence on any characteristic’s heritability, especially when multiple GWAS’s refuse to turn up significant evidence in favor of genetic contributions.

Over the past 200 years, the average male German’s height has increased by 7.6%. If height were truly 80-90% heritable, a 7.6% regional change — in the absence of genetic changes — is astonishing. Perhaps humanity is, as a whole, approaching our genetic potential for height. Or, perhaps, there are far more environmental factors at play.

Simeone et al. explain the disparity well (inline citations have been excerpted):

“Intriguingly, although, up to 90% of the variation in adult height may be explained by genetic factors, stature-associated polymorphisms have been found to only explain between 2% and 3.7% of height variation. More recent analyses have increased the combined predictive power of the identified traits. Nevertheless, a large fraction of heritable height‐associated factors has escaped detection by conventional GWAS, consistent with difficulties of previous association studies in finding variants robustly associated with height.”

Perhaps researchers simply need to keep looking for the golden egg of genetics. Or, perhaps, it’s just a wild goose chase.

Statistics & The GWAS

The GWAS has also amassed a certain amount of criticism for other shortcomings, as one article in the Journal of the American Medical Association explains.

“… The massive number of statistical tests performed presents an unprecedented potential for false-positive results, leading to new stringency in acceptable levels of statistical significance and requirements for replication of findings.”

Indeed, in practice, many genetics studies fall fate to the same issues Pearson, et al. point out. In some instances, the errors even cross the line into so-called “data dredging,” an abuse of statistics where a relationship can be found to be statistically significant despite no apparent causality.

In 2011, editors from the Behavior Genetics journal issued a warning regarding these studies:

“The literature on candidate gene associations is full of reports that have not stood up to rigorous replication … It now seems likely that many of the published findings of the last decade are wrong or misleading and have not contributed to real advances in knowledge.”

This isn’t to say that the GWAS isn’t entirely useless — on the contrary, it has provided some great advancements towards a more complete understanding of the human genome. But even these great advancements come with serious shortcomings, further casting doubt on the GWAS’s ability to identify genetic influences.

One GWAS performed on 16,175 women found seven SNPs associated with breast size (in the context of breast cancer research). The researchers noted a prior twin study which established that 56% of breast size is hereditable.

The shortcoming here is the confounding and lack of control for other variables attributing to breast size. A casual interpretation of the 56% heredity implies significant genetic expression for breast size. That conclusion is misleading as it ignores the covariate of body size.

About one-third of that 56% breast size heredity is “in common with genes influencing body mass index,” Wade et al. found. Indeed, the GWAS had concluded that genes known for an obesity risk factor — as a larger BMI is associated with a larger breast size — contributed to breast size. While that conclusion isn’t false, it’s certainly misleading, especially in the context of the study.

GWAS results are often rife with confounding variables like this, especially when a study’s authors don’t bother to control for certain variables.

This isn’t to say that the GWAS hasn’t seriously contributed to genetics research; on the contrary, it has allowed for important discoveries linking diseases with individual genes. But one must look critically when presented with data from a GWAS. Large sample sizes don’t fix observational studies. Studies ripe with confounding and p-hacking continue to make their way into reputable journals.

Even then, when the GWAS fails to identify a genetic basis for heritability, people routinely point to the omnigenetic model as an explanation. If the trait has no readily available explanation, it must be genetic, they say — even if the scientific method has no room for such a conjecture.

We’re all for science done right, but abuse of the GWAS is an example of science done wrong. For editors of prestigious medical journals to call their own published articles “wrong or misleading,” geneticists may have taken a serious misstep in the use of such a powerful tool.