You are currently browsing the tag archive for the ‘GWAS’ tag.

If you haven’t heard about Clubhouse yet… well, it’s the latest Silicon Valley unicorn, and the popular new chat hole for thought leaders. I heard about it for the first time a few months ago, and was kindly offered an invitation (Club house is invitation only!) so I could explore what it is all about. Clubhouse is an app for audio based social networking, and the content is, as far as I can tell, a mixed bag. I’ve listened to a handful of conversations hosted on the app.. topics include everything from bitcoin to Miami. It was interesting, at times, to hear the thoughts and opinions of some of the discussants. On the other hand, there is a lot of superficial rambling on Clubhouse as well. During a conversation about genetics I heard someone posit that biology has a lot to learn from the fashion industry. This was delivered in a “you are hearing something profound” manner, by someone who clearly knew nothing about either biology or the fashion industry, which is really too bad, because the fashion industry is quite interesting and I wouldn’t be surprised at all if biology has something to learn from it. Unfortunately, I never learned what that is.

One of the regulars on Clubhouse is Noor Siddiqui. You may not have heard of her; in fact she is officially “not notable”. That is to say, she used to have a Wikipedia page but it was deleted on the grounds that there is nothing about her that indicates notability, which is of course notable in and of itself… a paradox that says more about Wikipedia’s gatekeeping than Siddiqui (Russell 1903, Litt 2021). In any case, Siddiqui was recently part of a Clubhouse conversation on “convergence of genomics and reproductive technology” together with Carlos Bustamante (advisor to cryptocurrency based Luna DNA and soon to be professor of business technology at the University of Miami) and Balaji Srinivasan (bitcoin angel investor and entrepreneur). As it happens, Siddiqui is the CEO of a startup called “Orchid Health“, in the genomics and reproductive technology “space”. The company promises to harness “population genetics, statistical modeling, reproductive technologies, and the latest advances in genomic science” to “give parents the option to lower a future child’s genetic risk by creating embryos through in IVF and implanting embryos in the order that can reduce disease risk.” This “product” will be available later this year. Bustamante and Srinivasan are early “operators and investors” in the venture.

Orchid is not Siddiqui’s first startup. While she doesn’t have a Wikipedia page, she does have a website where she boasts of having (briefly) been a Thiel fellow and, together with her sister, starting a company as a teenager. The idea of the (briefly in existence) startup was apparently to help the now commercially defunct Google Glass gain acceptance by bringing the device to the medical industry. According to Siddiqui, Orchid is also not her first dive into statistical modeling or genomics. She notes on her website that she did “AI and genomics research”, specifically on “deep learning for genomics”. Such training and experience could have been put to good use but…

Polygenic risk scores and polygenic embryo selection

Orchid Health claims that it will “safely and naturally, protect your baby from diseases that run in your family” (the slogan “have healthy babies” is prominently displayed on the company’s website). The way it will do this is to utilize “advances in machine learning and artificial intelligence” to screen embryos created through in-vitro fertilization (IVF) for “breast cancer, prostate cancer, heart disease, atrial fibrillation, stroke, type 2 diabetes, type 1 diabetes, inflammatory bowel disease, schizophrenia and Alzheimer’s“. What this means in (a statistical geneticist’s) layman’s terms is that Orchid is planning to use polygenic risk scores derived from genome-wide association studies to perform polygenic embryo selection for complex diseases. This can be easily unpacked because it’s quite a simple proposition, although it’s far from a trivial one- the statistical genetics involved is deep and complicated.

First, a single-gene disorder is a health problem that is caused by a single mutation in the genome. Examples of such disorders include Tay-Sachs disease, sickle cell anaemia, Huntington’s disease, Duchenne muscular dystrophy, and many other diseases. A “complex disease”, also called a multifactorial disease, is a disease that has a genetic component, but one that involves multiple genes, i.e. it is not a single-gene disorder. Crucially, complex diseases may involve effects of environmental factors, whose role in causing disease may depend on the genetic composition of an individual. The list of diseases on Orchid’s website, including breast cancer, prostate cancer, heart disease, atrial fibrillation, stroke, type 2 diabetes, type 1 diabetes, inflammatory bowel disease, schizophrenia and Alzheimer’s disease are all examples of complex (multifactorial) diseases.

To identify genes that associate with a complex disease, researchers perform genome-wide association studies (GWAS). In such studies, researchers typically analyze several million genomic sites in a large numbers of individuals with and without a disease (used to be thousands of individuals, nowadays hundreds of thousands or millions) and perform regressions to assess the marginal effect at each locus. I italicized the word associate above, because genome-wide association studies do not, in and of themselves, point to genomic loci that cause disease. Rather, they produce, as output, lists of genomic loci that have varying degrees of association with the disease or trait of interest.

Polygenic risk scores (PRS), which the Broad Institute claims to have discovered (narrator: they were not discovered at the Broad Institute), are a way to combine the multiple genetic loci associated with a complex disease from a GWAS. Specifically, a PRS \hat{S} for a complex disease is given by

\hat{S} = \sum_{j=1}^m X_j \hat{\beta}_j,

where the sum is over m different genetic loci, the X_j are coded genetic markers for an individual at the m loci, and the \hat{\beta}_j are weights based on the marginal effects derived from a GWAS. The concept of a PRS is straightforward, but the details are complicated, in some cases subtle, and generally non-trivial. There is debate over how many genomic loci should be used in computing a polygenic risk score given that the vast majority of marginal effects are very close to zero (Janssens 2019), lots of ongoing research about how to set the weights to account for issues such as bias caused by linkage disequilibrium (Vilhjálmsson et al. 2015, Shin et al. 2017, Newcombe et al. 2019, Ge et al. 2019, Lloyd-Jones et al. 2019, Pattee and Pan 2020, Song et al. 2020), and continuing discussions about the ethics of using polygenic risk scores in the clinic (Lewis and Green 2021).

While much of the discussion around PRS applications centers on applications such as determining diagnostic testing frequency (Wald and Old 2019), polygenic embryo selection (PES) posits that polygenic risk scores should be taken a step further and evaluated for embryos to be used as a basis for discarding, or selecting, specific embryos for in vitro fertilization implantation. The idea has been widely criticized and critiqued (Karavani et al. 2019). It has been described as unethical, morally repugnant, and concerns about its use for eugenics have been voiced by many. Underlying these criticisms is the fact that the technical issues with PES using PRS are manifold.

Poor penetrance

The term “penetrance” for a disease refers to the proportion of individuals with a particular genetic variant that have the disease. Many single-gene disorders have very high penetrance. For example, F508del mutation in the CFTR gene is 100% penetrant for cystic fibrosis. That is, 100% of people who are homozygous for this variant, meaning that both copies of their DNA have a deletion of the phenylalanine amino acid in position 508 of their CFTR gene, will have cystic fibrosis. The vast majority of variants associated with complex diseases have very low penetrance. For example, in schizophrenia, the penetrance of “high risk” de novo copy number variants (in which there are variable copies of DNA at a genomic loci) was found to be between 2% and 7.4% (Vassos et al 2010). The low penetrance at large numbers of variants for complex diseases was precisely the rationale for developing polygenic risk scores in the first place, the idea being that while individual variants yield small effects, perhaps in (linear) combination they can have more predictive power. While it is true that combining variants does yield more predictive power for complex diseases, unfortunately the accuracy is, in absolute terms, very low.

The reason for low predictive power of PRS is explained well in (Wald and Old 2020) and is illustrated for coronary artery disease (CAD) in (Rotter and Lin 2020):

The issue is that while the polygenic risk score distribution may indeed be shifted for individuals with a disease, and while this shift may be statistically significant resulting in large odds ratios, i.e. much higher relative risk for individuals with higher PRS, the proportion of individuals in the tail of the distributions who will or won’t develop the disease will greatly affect the predictive power of the PRS. For example, Wald and Old note that PRS for CAD from (Khera et al. 2018) will confer a detection rate of only 15% with a false positive rate of 5%. At a 3% false positive rate the detection rate would be only 10%. This is visible in the figure above, where it is clear that control of the false positive right (i.e. thresholding at the extreme right-hand side with high PRS score) will filter out many (most) affected individuals. The same issue is raised in the excellent review on PES of (Lázaro-Muńoz et al. 2020). The authors explain that “even if a PRS in the top decile for schizophrenia conferred a nearly fivefold increased risk for a given embryo, this would still yield a >95% chance of not developing the disorder.” It is worth noting in this context, that diseases like schizophrenia are not even well defined phenotypically (Mølstrøm et al. 2020), which is another complex matter that is too involved to go into detail here.

In a recent tweet, Siddiqui describes natural conception as a genetic lottery, and suggests that Orchid Health, by performing PES, can tilt the odds in customers’ favor. To do so the false positive rate must be low, or else too many embryos will be discarded. But a 15% sensitivity is highly problematic considering the risks inherent with IVF in the first place (Kamphuis et al. 2014):

To be concrete, an odds ratio of 2.8 for cerebral palsy needs to be balanced against the fact that in the Khera et al. study, only 8% of individuals had an odds ratio >3.0 for CAD. Other diseases are even worse, in this sense, than CAD. In atrial fibrillation (one of the diseases on Orchid Health’s list), only 9.3% of the individuals in the top 0.44% of the atrial fibrillation PRS actually had atrial fibrillation (Choi et al 2019).As one starts to think carefully about the practical aspects and tradeoffs in performing PES, other issues, resulting from the low penetrance of complex disease variants, come into play as well. (Lencz et al. 2020) examine these tradeoffs in detail, and conclude that “the differential performance of PES across selection strategies and risk reduction metrics may be difficult to communicate to couples seeking assisted reproductive technologies… These difficulties are expected to exacerbate the already profound ethical issues raised by PES… which include stigmatization, autonomy (including “choice overload”, and equity. In addition, the ever-present specter of eugenics may be especially salient in the context of the LRP (lowest-risk prioritization) strategy.” They go on to “call for urgent deliberations amongst key stakeholders (including researchers, clinicians, and patients) to address governance of PES and for the development of policy statements by professional societies.”

Pleiotropy predicaments

I remember a conversation I had with Nicolas Bray several years ago shortly after the exciting discovery of CRISPR/Cas9 for genome editing, on the implications of the technology for improving human health. Nick pointed out that the development of genomics had been curiously “backwards”. Thirty years ago, when human genome sequencing was beginning in earnest, the hope was that with the sequence at hand we would be able to start figuring out the function of genes, and even individual base pairs in the genome. At the time, the human genome project was billed as being able to “help scientists search for genes associated with human disease” and it was imagined that “greater understanding of the genetic errors that cause disease should pave the way for new strategies in diagnosis, therapy, and disease prevention.” Instead, what happened is that genome editing technology has arrived well before we have any idea of what the vast majority of the genome does, let alone the implications of edits to it. Similarly, while the coupling of IVF and genome sequencing makes it possible to select embryos based on genetic variants today, the reality is that we have no idea how the genome functions, or what the vast majority of genes or variants actually do.

One thing that is known about the genome is that it is chock full of pleiotropy. This is statistical genetics jargon for the fact that variation at a single locus in the genome can affect many traits simultaneously. Whereas one might think naïvely that there are distinct genes affecting individual traits, in reality the genome is a complex web of interactions among its constituent parts, leading to extensive pleiotropy. In some cases pleiotropy can be antagonistic, which means that a genomic variant may simultaneously be harmful and beneficial. A famous example of this is the mutation to the beta globin gene that confers malaria resistance to heterozygotes (individuals with just one of their DNA copies carrying the mutation) and sickle cell anemia to homozygotes (individuals with both copies of their DNA carrying the mutation).

In the case of complex diseases we don’t really know enough, or anything, about the genome to be able to truly assess pleiotropy risks (or benefits). But there are some worries already. For example, HLA Class II genes are associated with Type I and non-insulin treated Type 2 diabetes (Jacobi et al 2020), Parkinson’s disease (e.g. James and Georgopolous 2020, which also describes an association with dementia) and Alzheimer’s (Wang and Xing 2020). PES that results in selection against the variants associated with these diseases could very well lead to population susceptibility to infectious disease. Having said that, it is worth repeating that we don’t really know if the danger is serious, because we don’t have any idea what the vast majority of the genome does, nor the nature of antagonistic pleiotropy present in it. Almost certainly by selecting for one trait according to PRS, embryos will also be selected for a host of other unknown traits.

Thus, what can be said is that while Orchid Health is trying to convince potential customers to not “roll the dice“, by ignoring the complexities of pleiotropy and its implications for embryo selection, what the company is actually doing is in fact rolling the dice for its customers (for a fee).

Population problems

One of Orchid Health’s selling points is that unlike other tests that “look at 2% of only one partner’s genome…Orchid sequences 100% of both partner’s genomes” resulting in “6 billion data points”. This refers to the “couples report”, which is a companion product of sorts to the polygenic embryo screening. The couples report is assembled by using the sequenced genomes of parents to simulate the genomes of potential babies, each of which is evaluated for PRS’ to provide a range of (PRS based) disease predictions for the couples potential children. Sequencing a whole genome is a lot more expensive that just assessing single nucleotide polymorphisms (SNPs) in a panel. That may be one reason that most direct-to-consumer genetics is based on polymorphism panels rather than sequencing. There is another: the vast majority of variation in the genome occurs at a known polymorphic sites (there are a few million out of the approximately 3 billion base pairs in the genome), and to the extent that a variant might associate with a disease, it is likely that a neighboring common variant, which will be inherited together with the causal one, can serve as a proxy. There are rare variants that have been shown to associate with disease, but whether or not they explain can explain a large fraction of (genetic) disease burden is still an open question (Young 2019). So what has Siddiqui, who touts the benefits of whole-genome sequencing in a recent interview, discovered that others such as 23andme have missed?

It turns out there is value to whole-genome sequencing for polygenic risk score analysis, but it is when one is performing the genome-wide association studies on which the PRS are based. The reason is a bit subtle, and has to do with differences in genetics between populations. Specifically, as explained in (De La Vega and Bustamante, 2018), variants that associate with a disease in one population may be different than variants that associate with the disease in another population, and whole-genome sequencing across populations can help to mitigate biases that result when restricting to SNP panels. Unfortunately, as De La Vega and Bustamante note, whole-genome sequencing for GWAS “would increase costs by orders of magnitude”. In any case, the value of whole-genome sequencing for PRS lies mainly in identifying relevant variants, not in assessing risk in individuals.

The issue of population structure affecting PRS unfortunately transcends considerations about whole-genome sequencing. (Curtis 2018) shows that PRS for schizophrenia is more strongly associated with ancestry than with the disease. Specifically, he shows that “The PRS for schizophrenia varied significantly between ancestral groups and was much higher in African than European HapMap subjects. The mean difference between these groups was 10 times as high as the mean difference between European schizophrenia cases and controls. The distributions of scores for African and European subjects hardly overlapped.” The figure from Curtis’ paper showing the distribution of PRS for schizophrenia across populations is displayed below (the three letter codes at the bottom are abbreviations for different population groups; CEU stands for Northern Europeans from Utah and is the lowest).

The dependence of PRS on population is a problem that is compounded by a general problem with GWAS, namely that Europeans and individuals of European descent have been significantly oversampled in GWAS. Furthermore, even within a single ancestry group, the prediction accuracy of PRS can depend on confounding factors such as socio-economic status (Mostafavi et al. 2020). Practically speaking, the implications for PES are beyond troubling. The PRS scores in the reports customers of Orchid Health may be inaccurate or meaningless due to not only the genetic background or admixture of the parents involved, but also other unaccounted for factors. Embryo selection on the basis of such data becomes worse than just throwing dice, it can potentially lead to unintended consequences in the genomes of the selected embryos. (Martin et al. 2019) show unequivocally that clinical use of polygenic risk scores may exacerbate health disparities.

People pathos

The fact that Silicon Valley entrepreneurs are jumping aboard a technically incoherent venture and are willing to set aside serious ethical and moral concerns is not very surprising. See, e.g. Theranos, which was supported by its investors despite concerns being raised about the technical foundations of the company. After a critical story appeared in the Wall Street Journal, the company put out a statement that

“[Bad stories]…come along when you threaten to change things, seeded by entrenched interests that will do anything to prevent change, but in the end nothing will deter us from making our tests the best and of the highest integrity for the people we serve, and continuing to fight for transformative change in health care.”

While this did bother a few investors at the time, many stayed the course for a while longer. Siddiqui uses similar language, brushing off criticism by complaining about paternalism in the health care industry and gatekeeping, while stating that

“We’re in an age of seismic change in biotech – the ability to sequence genomes, the ability to edit genomes, and now the unprecedented ability to impact the health of a future child.”

Her investors, many of whom got rich from cryptocurrency trading or bitcoin, cheer her on. One of her investors is Brian Armstrong, CEO of Coinbase, who believes “[Orchid is] a step towards where we need to go in medicine.” I think I can understand some of the ego and money incentives of Silicon Valley that drive such sentiment. But one thing that disappoints me is that scientists I personally held in high regard, such as Jan Liphardt (associate professor of Bioengineering at Stanford) who is on the scientific advisory board and Carlos Bustamante (co-author of the paper about population structure associated biases in PRS mentioned above) who is an investor in Orchid Health, have associated themselves with the company. It’s also very disturbing that Anne Wojcicki, the CEO of 23andme whose team of statistical geneticists understand the subtleties of PRS, still went ahead and invested in the company.

Conclusion

Orchid Health’s polygenic embryo selection, which it will be offering later this year, is unethical and morally repugnant. My suggestion is to think twice before sending them three years of tax returns to try to get a discount on their product.

The Bulbophyllum echinolabium orchid. The smell of its flowers has been described as decomposing rot.

[Update April 6, 2014: The initial title of this post was “23andme genotypes are all wrong”. While that was and remains a technically correct statement, I have changed it because the readership of my blog, and this post in particular, has changed. Initially, when I made this post, the readers of the blog were (computational) biologists with extensive knowledge of genotyping and association mapping, and they could understand the point I was trying to make with the title. However in the past few months the readership of my blog has grown greatly, and the post is now reaching a wide public audience. The revised title clarifies that the content of this post relates to the point that low error rates in genotyping can be problematic in the context of genome-wide association reports because of multiple-testing.]

I have been reading the flurry of news articles and blog posts written this week about 23andme and the FDA with some interest. In my research talks, I am fond of displaying 23andme results, and have found that people always respond with interest. On the teaching side, I have subsidized 23andme testing for volunteer students in Math127 who were interested in genetics so that they could learn about personalized genomics first-hand. Finally, a number of my former and current students have worked at 23andme, and some are current employees.

Despite lots of opinions being expressed about the 23andme vs. FDA kerfuffle, I believe that two key points have been ignored in the discussions:

  1. All 23andme genotypes that have ever been reported to customers are wrong. This is the case despite very accurate genotyping technology used by 23andme.
  2. The interpretation of 23andme results involves examining a large number of odds ratios. The presence of errors leads to a huge multiple-testing problem.

Together, these issues lead to an interesting conundrum for the company, for customers, and for the FDA.

I always find it useful to think about problems concretely. In the case of 23andme, it means examining actual genotypes. Fortunately, you don’t have to pay the company $99 dollars to get your own- numerous helpful volunteers have posted their 23andme genotypes online. They can be viewed at openSNP.org where “customers of direct-to-customer genetic tests [can] publish their test results, find others with similar genetic variations, learn more about their results, get the latest primary literature on their variations and help scientists find new associations”. There are a total of 624 genotypes available at openSNP, many of them from 23andme. As an example, consider “samantha“, who in addition to providing her 23andme genotype, also provides lots of phenotypic information. Here is the initial part of her genotype file:

# This data file generated by 23andMe at: Wed Jul 20 20:37:11 2011
#
# Below is a text version of your data. Fields are TAB-separated
# Each line corresponds to a single SNP.  For each SNP, we provide its identifier 
# (an rsid or an internal id), its location on the reference human genome, and the 
# genotype call oriented with respect to the plus strand on the human reference 
# sequence.     We are using reference human assembly build 36.  Note that it is possible 
# that data downloaded at different times may be different due to ongoing improvements 
# in our ability to call genotypes. More information about these changes can be found at:
# https://www.23andme.com/you/download/revisions/
# 
# More information on reference human assembly build 36:
# http://www.ncbi.nlm.nih.gov/projects/mapview/map_search.cgi?taxid=9606&build=36
#
# rsid	chromosome	position	genotype
rs4477212	1	72017	AA
rs3094315	1	742429	AG
rs3131972	1	742584	AG
rs12124819	1	766409	AA
rs11240777	1	788822	AA
rs6681049	1	789870	CC
rs4970383	1	828418	CC
rs4475691	1	836671	CC
rs7537756	1	844113	AA
rs13302982	1	851671	GG
rs1110052	1	863421	GT
...

Anyone who has been genotyped by 23andme can get this file for themselves from the website (by clicking on their name, then on “Browse Raw Data” from the pull-down menu, and then clicking on “Download” in the top-right corner of the browser window). The SNPs are labeled with rsid labels (e.g. rs3094315) and correspond to specific locations on chromosomes (e.g. chr1:742429). Since every human is diploid, two bases are shown for every SNP; one came from mom and one from dad. The 23andme genotype is not phased, which means that you can’t tell in the case of rs3094315 whether the A was from mom and the G from dad, or vice versa (it turns out paternal origin can be important, but that is a topic for another post).

A key question the FDA has asked, as it does for any diagnostic test, is whether the SNP calls are accurate. The answer is already out there. First, someone has performed a 23andme replicate experiment precisely to assess the error rate. In an experiment in 2010 with two replicates, 85 SNPs out of about 600,000 were different. Today, Illumina types around 1 million SNPs, so one would expect even more errors. Furthermore, a replicate analysis provides only a lower bound, since systematic errors will not be detected. Another way to examine the error rate is to look at genotypes of siblings. That was written about in this blog post which concluded there were 87 errors. 23andme currently uses the Illumina Omni Express for genotyping, and the Illumina spec sheet claims a similar error rate to those inferred in the blog posts mentioned above. The bottom line is that even though the error rate for any individual SNP call is very very low (<0.01% error), with a million SNPs being called there is (almost) certainly at least one error somewhere in the genotype. In fact, assuming a conservative error rate leading to an average of 100 errors per genotype, the probability that a 23andme genotype has no errors is less than 10^(-40).

The fact that 23andme genotypes are wrong (i.e. at least one error in some SNP) wouldn’t matter if one was only interested in a single SNP. With very high probability, it would be some other SNPs that are the wrong ones. But the way people use 23andme is not to look at a single SNP of interest, but rather to scan the results from all SNPs to find out whether there is some genetic variant with large (negative) effect. The good news is that there isn’t much information available for the majority of the 1 million SNPs being tested. But there are, nevertheless, lots of SNPs (thousands) to look at. Whereas a comprehensive exam at a doctor’s office might currently constitute a handful of tests– a dozen or a few dozen at most– a 23andme test assessing thousands of SNPs and hundreds of diseases/traits constitutes more diagnostic tests on an individual at one time than have previously been performed in a lifetime.

To understand how many tests are being performed in a 23andme experiment, it is helpful to look at the Interpretome website. The website allows a user to examine information on SNPs without paying, and without uploading the data. I took a look at Samantha, and the Interpretome gave information about 2829 SNPs. These are SNPs for which there is a research article that has identified the SNP as significant in some association study (the website conveniently provides direct links to the articles). For example, here are two rows from the phenotype table describing something about Samantha’s genetic predisposition for large head circumference:

Head circumference (infant) 11655470  CC  T  .05    4E-6    22504419
Head circumference (infant) 1042725    CC  T  .07    3E-10  22504419

Samantha’s genotype at the locus is CC, the “risk” allele is T, the odds ratios  are very small (0.05,0.07) and the p-values are apparently significant. Interpretome’s results differ from those of 23andme, but looking at the diversity of phenotypes reported on gives one a sense for the possibilities that currently exist in genetics, and the scope of 23andme’s reports.

From the estimates of error rates provided above, and using the back of an envelope, it stands to reason that about 1/3 of 23andme tested individuals have an error at one of their “interesting” SNPs. Not all of SNPs arising in association studies are related to diseases, but many of them are. I don’t think its unreasonable to postulate that a significant percentage of 23andme customers have some error in a SNP that is medically important. Whether such errors are typically false positives or false negatives is unclear, and the extent to which they may lead to significant odds ratios is another interesting question. In other words, its not good enough to know how frequently warfarin sensitivity is being called incorrectly. The question is how frequently some medically significant result is incorrect.

Of course, the issue of multiple testing as it pertains to interpreting genotypes is probably a secondary issue with 23andme. As many bloggers have pointed out, it is not even clear that many of 23andme’s odds ratios are accurate or meaningful. A major issue, for example, is the population background of an individual examining his/her genotype and how close it is to the population on which the GWAS were performed. Furthermore, there are serious questions about the meaning of the GWAS odds ratios in the case of complex traits. However I think the issue of multiple testing is a deeper one, and a problem that will only be exacerbated as more disease SNPs are identified. Having said that, there are also approaches that could mitigate errors and improve fidelity of the tests. As DECODE genetics has demonstrated, imputation and phasing can in principle be used to infer population haplotypes, which not only are useful for GWAS analyses, but can also be used to identify erroneous SNP calls. 23andme’s problem is that although they have many genotypes, they are from diverse populations that will be harder to impute and phase.

The issue of multiple testing arising in the context of 23andme and the contrast with classic diagnostics reminds me of the dichotomy between whole-genome analysis and classic single gene molecular biology. The way in which customers are looking at their 23andme results is precisely to look for the largest effects, i.e. phenotypes where they appear to have high odds of contracting a disease, or being sensitive to some drug. This is the equivalent of genome scientists picking the “low hanging fruit” out of genome-wide experiments such as those performed in ENCODE. In genomics, scientists have learned (with some exceptions)  how to interpret genome-wide analyses after correcting for multiple-hypothesis testing by controlling for false discovery rate. But are the customers of 23andme doing so? Is the company helping them do it? Should it? Will the FDA require it? Can looking at ones own genotype constitute too much testing?

There are certainly many precedents for superfluous harmful testing in medicine. For example, the American Academy of Family Physicians has concluded that prostate cancer PSA tests and digital rectal exams have marginal benefits that are outweighed by the harm caused by following up on positive results. Similar arguments have been made for mammography screening. I therefore think that there are serious issues to consider about the implications of direct-to-consumer genetic testing and although I support the democratization of genomics, I’m glad the FDA is paying attention.

samantha

Samantha’s type 2 diabetes risk as estimated from her genotype by Interpretome. She appears to have a lower risk than an average person. Does this make it ok for her to have another cookie?

Blog Stats

  • 2,657,677 views
%d bloggers like this: