Rare Diseases

This week we celebrate Rare Disease Day, which aims to raise awareness for rare diseases and improve the medical and research support for individuals with rare diseases and their families. I’ve worked in the rare disease research space for several years, and it does feel like the importance of these conditions is increasingly recognized. Earlier this week, U.S. Senators Sherrod Brown (D-OH) and John Barrasso (R-WY) announced the passage of their Senate resolution designating February 28, 2023 as “Rare Disease Day” in the US. When the two major political parties in our country agree on something, it’s kind of a big deal.

Our institution, like many others, hosted an event for patients, families, clinicians, and researchers who are invested in rare disease. It was a special occasion this year as we returned to in-person and celebrated the recognition of Nationwide Children’s Hospital as a Rare Disease Center of Excellence by the National Organization for Rare Disorders (NORD). The keynote speaker was Dr. Jerry Mendell, a pioneer in gene therapy, who shared some of the challenges and successes of developing viral gene therapies for devastating diseases like spinal muscular atrophy and Duchenne Muscular Dystrophy. Watching videos of his patients before and after therapy… well, it’s inspiring to say the least.

The State of Rare Disease Research in 2023

The first step to treating a patient with a rare disease is to obtain a molecular diagnosis. Because the vast majority of rare diseases have genetic origins, this is an area that has seen dramatic advances in the era of high-throughput sequencing. I listen to many scientific talks about disease genetics and gene discovery. Many of them report around ~8,000 gene-disease associations currently in the OMIM database. This week I heard that modern counts are higher, something closer to 12,000. Anecdotally, it’s obvious to those of us who work in gene discovery that at least 2-3 new disease-gene relationships are published in peer-reviewed journals per week.

For many clinical and research laboratories, the OMIM Database is the primary authority on human gene disease associations. It’s a curated resource that is free to the public and maintained by Johns Hopkins. A few months ago, I downloaded the entire database of OMIM associations and did some rudimentary analyses. At that time, OMIM contained:

• 18,077 recognized human genes
• 7,333 phenotypes mapped to a gene
• 6,473 of which are genetic disorders (versus susceptibility and quantitative trait associations)
• 5,955 of which are Mendelian disorders with an established pattern of inheritance

Inheritance of OMIM Disease Genes

As you can see from the pie chart, autosomal recessive disorders are the most prevalent category (53%), followed by autosomal dominant (35%). I imagine the gap between those two categories has been shrinking, as the majority of newly discovered genes in this era are associated with de novo mutations and thus classified as autosomal dominant. A small proportion of disorders are described as both AD and AR. Sometimes that is true. Other times, most patients in the literature are biallelic (AR), but there’s some [usually 10+ years old] published study of a patient with disease who has only one heterozygous variant in the gene. In my experience, if that patient were studied with modern approaches — whole-genome sequencing and transcriptome analysis — there’s a good chance a second variant would be identified.

X-linked inheritance patterns were lumped together for the pie chart, but most are X-linked recessive (n=216), as opposed to X-linked dominant (n=62) or unspecified X-linked inheritance (n=69). With the latter category, if you go and read about those disorders, many have only been reported in affected males and thus are likely to be X-linked recessive.

Somatic inheritance is not really a Mendelian pattern, but I included it as the 5th most common category of OMIM disease genes with >200 curated. Most of these are cancers or cancer syndromes associated with well-known genes like APC, KRAS, and PIK3CA, but there are a handful of somatic-mosaic disorders recognized. This is another category likely to grow substantially as our ability to detect disease-causing variation in relevant tissues and cell types improves.

Other inheritance patterns are recognized by OMIM but are too rare to be shown in a pie chart. These include things like digenic inheritance (n=16) and pseudo-autosomal inheritance (n=4). There are even a handful of Y-linked disorders. Frankly, I’m dubious of most associations in these categories, but once something is established in OMIM, it’s hard to go back.

Curation versus Gene Discovery

One of the major reasons OMIM is so popular is that it’s curated and updated with biomedical literature. In the genomics field, we have come to love curated resources because, generally speaking, the quality is much higher. Yet in the case of OMIM, it is clear that curation lags far behind the pace of discovery. This is unsurprising. New disease genes and expanded phenotypes are being published daily across a wide array of peer-reviewed journals. It’s hard for anyone to keep up.

Yet for patients with undiagnosed rare diseases, recent discoveries are critical. Multiple studies of the reanalysis of previously-negative exome tests have demonstrated that once 2-3 years have passed, a significant proportion (10-20%) yield diagnostic results when reanalyzed. Most of the time, it’s not pipeline improvements or sophisticated analytical tools, but due to the fact that the diagnostic finding is in a newly discovered disease gene, or a new phenotypic manifestation of a known disease gene.

Given the value of up-to-date information, how do we strike the balance? One possibility is to leverage crowdsourcing of experts in the field. This is what the Clinical Genome Resource (ClinGen) is doing with its expert panels, which interact with various working groups. I’m part of the Variant Curation Expert Panel for RPE65, a gene linked to autosomal recessive Leber Congenital Amaurosis (progressive blindness) — our duty is to help establish extremely rigorous guidelines for variant interpretation of variants in this gene. This goes a level beyond ACMG guidelines, which are fairly broad, and applies some gene/disease-specific rules for how each type of evidence could should be evaluated. The VCEP panel is complemented by a GCEP (Gene Curation Expert Panel) which reviews and establishes gene-disease associations.

Everyone on an expert panel is a volunteer, and each panel has member with complementary expertise. Since we all have a personal stake in diagnostics for this gene (or condition), we’re all invested in doing the work and reaching a useful consensus when necessary. I just took a look and wow, there are many GCEPs and VCEPs already established. That seems encouraging.

Rare Disease Genomics in 2023

There is plenty of work to do, of course. Many groups including ours will continue sequencing undiagnosed patients and connecting with other investigators via Matchmaking to help establish new disease genes. Trend-wise, I expect at least two new paradigms will emerge:

1. Leveraging large-scale biobanks with genomic data to validate new disease genes and identify additional patients to better define phenotypes.
2. Applying ever-more-sophisticated functional assays, including organoids, gene-edited model organisms, and other approaches, to support and explore the mechanisms of disease.

It’s going to be a busy year, but a good one, for Rare Diseases.

High-throughput sequencing has accelerated human genetics research in countless ways, from rapid newborn sequencing to Mendelian disorders to the many discoveries being made in common, complex disease cohorts. Yet one of the most fundamental benefits of this technology is our ability to catalogue human genetic variation in large cohorts of individuals. For a long while, such catalogues necessarily focused on variation in human protein-coding genes. Coding regions harbor the vast majority of medically-relevant sequence variants — in part because their biological effects are easier to predict — but they only occupy a small 1-2% of the human genome.

Whole-genome sequencing offers a far more comprehensive interrogation of genetic variation. The gnomAD database and more recently the Trans-Omics for Precision Medicine (TopMed) program have catalogued the genomes of tens of thousands of individuals. Unlike the 1,000 Genomes Project, which surveyed the genomes of numerous diverse/representative world populations without collecting any phenotypic data, most of the individuals in gnomAD and TopMed are part of disease studies. Even so, the sheer size of these cohorts has made them vital tools for human genetics in both clinical and research settings.

Origins and Composition of the gnomAD Database

The genome aggregation database (gnomAD), owing to is size and accessibility, has served as a vital resource for human genetics over the past several years. Like its exome-centric predecessor (ExAC), gnomAD’s catalogue was produced by compiling, harmonizing, and generating summary data across numerous large-scale sequencing projects. More than 60 projects contributed data to the gnomAD database. Most of these are disease cohorts such as TCGA (cancer), ADSP (Alzheimer’s), and T2D-GENES (diabetes). There are two major releases of gnomAD available with some key differences between them:

Release:	gnomAD v2.1.1	gnomAD v3.1.2
Assembly:	build37/hg19	GRCh38
Genomes:	15,708	76,156
Exomes:	125,748	0

As you’ll noticed, the first key difference is the genome assembly: gnomAD v2.1.1, which has data from more individuals but is exon-centric, is on build 37. The more recent release, v3.1.2, is on the newer assembly (GRCh38) but contains only genomes. This dichotomy is understandable; it’s a lot more work to harmonize exome sequencing data because it was generated using multiple version of target enrichment kits from several different manufacturers, all with a slightly different definition of exome. However, it’s also unfortunate because neither release contains the maximum available information. In fact, it’s an often-cited reason by labs for not moving to the newer* genome assembly (GRCh38).

*Newer is a relative term. GRCh38 was released in December 2013, more than 8 years ago.

The Caveats of gnomAD

The gnomAD database is a wonderful resource. I use it every day, as do many of my colleagues. Yet anyone who relies on gnomAD data for analysis and interpretation of genetic variants should be fully aware of its caveats.

1. Not everyone in gnomAD is healthy.

As I mentioned earlier, more than half of the contributing projects are disease studies (especially cardiovascular disease studies). Although individuals with severe pediatric disease and their first-degree relatives are excluded, “some individuals with severe disease may still be included in the data sets, albeit likely at a frequency equivalent to or lower than that seen in the general population.”

2. Not everyone in gnomAD is young

In fact, very few people are. According to a talk by the gnomAD team at ASHG 2017, the average age of a gnomAD individual is 54 years old. Maybe it should not have been given the types of disease cohorts that went into it, but it’s rather advanced, and brings up another important caveat.

3. Some variants are somatic clonal mutations

This came to attention a few years ago when researchers observed that gnomAD contained loss-of-function variants in key developmental genes that also happen to be frequently mutated in myelodysplastic syndrome.

A great example of this phenomenon is ASXL1, a gene that encodes a member of the Polycomb group of proteins which are necessary for the maintenance of stable repression of homeotic and other loci. De novo loss-of-function mutations in ASXL1 cause Bohring-Opitz syndrome, a severe congenital malformation disorder, so it was initially puzzling to observe nonsense/frameshift/splice site variants in some individuals in the gnomAD database. Especially severe truncating mutations that would be expected to cause BOS if they were present in an embryo.

Closer inspection of the aligned sequencing data for individuals carrying such variants indicates that they’re probably mosaic alterations, i.e. present in a subpopulation of cells. For example, two gnomAD individuals carry a variant at chr20-31021118-C-T (GRCh37) encodes a nonsense change (p.Gln373Ter) that’s Pathogenic in ClinVar. Scroll down on the variant page and you can see the aligned sequence data:

Genome sequencing is generally done on DNA extracted from blood, and the older a person is, the more likely they have mosaic hematopoietic cell populations. The proliferation advantage of ASXL1 mutations allows them to reach appreciable allele frequencies in blood cells. As a result, we observe a fair number of recurrent LOF variants — such as p.Arg417Ter (9 heterozygotes), p.Arg404Ter (7 heterozygotes), and p.Arg965Ter (3 heterozygotes) — all of which appear to be mosaic clonal events. The gnomAD team even added a note to the page for ASXL1:

Analysis of allele balance and age data indicates that this gene shows evidence of clonal hematopoiesis of indeterminate potential (CHIP). The potential presence of somatic variants should be taken into account when interpreting the penetrance, pathogenicity, and frequency of assumed germline variants. For more information, see pages 37-40 of supplementary information for The mutational constraint spectrum quantified from variation in 141,456 humans and Pathogenic ASXL1 somatic variants in reference databases complicate germline variant interpretation for Bohring-Opitz Syndrome.

This is a reason that the presence of a variant in gnomAD — even in a few individuals — should be considered carefully when evaluating its pathogenicity.

4. Many populations are underrepresented.

The individuals in gnomAD were not selected as representatives of world populations, but rather as the groups selected for large-scale sequencing projects. Here’s the breakdown by population for the two major versions of the database:

Unsurprisingly, the majority are of Western European ancestry. There are also a large number of Finnish individuals, whose unique population history makes them especially valuable for genetic studies. Yet many other significant world populations are under-represented.

Summary: gnomAD is great, use with appropriate caution

I’d intended to also talk about the TopMed database, but since I’m already a thousand words in I’ll have to save that for another time. In summary, the gnomAD database is a spectacular resource for human genetics. Like any resource, it is not without certain flaws. Anyone using it should be aware of these caveats and account for them in their analyses.

Some of the (many) recent advances in genetic testing are in the area of non-invasive prenatal testing, or NIPT. This form of genetic screening utilizes the blood of an expecting mother to screen for chromosomal abnormalities and other rare disorders in the fetus. It’s an area of intensive research at the moment, and also the subject of a high-profile article (probably paywalled) in the New York Times by Sarah Kliff and Aatish Bhatia. The story was published on New Year’s Day with the provocative title:

When They Warn of Rare Disorders, These Prenatal Tests Are Usually Wrong.

Essentially, it’s about the advent of NIPT in the United States, the companies who offer the tests, and the positive predictive value of the results. This excerpt captures the thrust of it:

In just over a decade, the tests have gone from laboratory experiments to an industry that serves more than a third of the pregnant women in America, luring major companies like Labcorp and Quest Diagnostics into the business, alongside many start-ups.

The tests initially looked for Down syndrome and worked very well. But as manufacturers tried to outsell each other, they began offering additional screenings for increasingly rare conditions.

The grave predictions made by those newer tests are usually wrong, an examination by The New York Times has found.

The NYT’s “examination” included, among other things:

• Pooling the results of five published studies on NIPT outcomes
• Conducting interviews with genetic counselors and other healthcare professionals
• Reviewing the marketing materials of several commercial NIPT offerings
• Collecting personal stories from patients who received false positive results

On the Accuracy of Rare Positive Results

Notice the subtle and careful wording used in the headline and in the excerpt above:

“When they warn of rare disorders”

“The grave predictions made”

In scientific terms, both of these phrases are referring to positive screening results, not all screening results. My response to the sensationalist claim that positive NIPT tests are often false positives? Of course they are. This is the nature of screening. I’m reminded of it every time I go through metal detectors at the airport.

The excerpt I provided is not the start of the article. This is the NYT and they know what they’re doing. The article opens with the story of a young pregnant woman whose NIPT returns a scary-sounding diagnosis for her unborn child (Prader-Willi syndrome) that turns out to be a false positive. It goes on to briefly summarize the rise of NIPT and its expansion to include increasingly rare disorders, especially ones associated with microdeletions.

The outcome, as illustrated by numerous graphics like the one at right, is that for every 15 times such tests correctly identify a genomic alteration, they are wrong 85 times. However, what none of these visually striking graphics do tell you is that for every positive result there are thousands of negative results. In other words, the overall accuracy of NIPT is extremely high.

Genomic Alteration Detection 101: Size Matters

Chromosomal abnormalities are a common cause of syndromic birth defects. The most prevalent of these, trisomy 21 (the cause of Down syndrome), has a prevalence of 1 in 664 newborns according to Smith’s Recognizable Patterns of Human Malformation (8th edition). Naturally, whole-chromosome or chromosome-arm abnormalities are also among the easiest things to detect by NIPT. Even with the technical challenges involved, such events leave massive footprints in the genome. That combination — large size and relative prevalence — is why early forms of NIPT that screened for trisomy 21 performed very well.

The term microdeletion comes from the field of cytogenetics and it’s a bit deceptively named: it refers to genomic deletions that are too small to be identified by looking at chromosomes with light microscopy. Yes, these events are more challenging to detect by NIPT than losses or gains of entire chromosomes. Even so, such microdeletions can still be huge, i.e. millions of base pairs long, and encompass dozens or hundreds of genes. In the prenatal setting, they tend to cause severe syndromic disorders. Many such disorders are extremely rare, and as noted, they can be more challenging to detect with accuracy. So yes, false positives may be more likely. However, something glossed over rather quickly by the NYT reporters is this: a positive screening result is not a diagnosis. It’s a signal that more testing should be performed.

Population Screening Benefits and Consequences

In biology in general, false positives for extremely rare things are expected because of the signal-to-noise ratio. A classic example that illustrates this is the identification of inherited versus de novo variants in a child.

Every human carries about 4-6 million sequence variants relative to the reference sequence. The vast majority of those are inherited from one’s parents. Most, in fact, are common in human populations because they arose a long time ago. In contrast, mutations that arise de novo in a child (i.e. are absent from the parents) occur at a rate of about 1e-08, making them extremely rare. On average, there are 50-70 such mutations genome-wide. When we sequence a family trio, we identify inherited variants with extremely high accuracy, i.e. >99.9%. Sure, there are a few hundred false positives, but there are many million true positives. The signal-to-noise ratio is very high. This changes when we look only for de novo mutations. Now it’s 50 true positives, 200 false positives. Very different signal to noise ratio.

However, and this is important, we don’t blithely call 200 de novo mutations in a child and assume they’re all real. With further scrutiny (e.g. using population databases and removing common artifacts) we can filter out most false positives and get to the correct number.

Positive results from NIPT also get follow-up. In the case of the woman introduced at the start of the NYT article, an amniocentesis later revealed that the fetus did not have Prader-Willi syndrome. And this, perhaps, is the biggest thing missed by the reporters: positive NIPT results are the beginning of a process. Many of those frightening positive results are later refuted by direct molecular testing or imaging studies. This does not mean we should stop screening altogether or demand that the screen yield perfect results. If it did, we wouldn’t do PSA screens or mammograms.

It’s like my airport analogy: most people who set off a metal detector are not carrying a weapon. That’s why TSA agents don’t open fire at the sound of a buzzer. When I set one off — which seems to be one of my talents in life — I empty my pockets and try again. Yet I don’t want them to stop checking people with metal detectors.

Emotional Trauma of False Positives

This is not to discount or ignore the emotional trauma that a positive screening result brings. These are very real consequences for the families affected. If you’re reading this, you’re probably close to someone who got a scary-sounding result of a medical test (as I am). Even if unconcerned, a possible diagnosis is usually terrifying.

Medical professionals take this risk quite seriously. It’s one of the major topics considered whenever population screening is discussed, one of the “costs” in a cost-benefit analysis. Yet we still do a lot of population screening because the benefits of early detection for many conditions outweigh the potential consequences.

Summary and Outlook for NIPT

I read the New York Times regularly, and I appreciate that they’re one outlet that produces in-depth articles, often about scientific/technical topics like genomics, which are backed by real reporting. I also acknowledge a lesson given to me by my AP English teacher in high school: good writing sometimes needs to take a stance on something. Yet I think this particular article has far too much of a negative slant. It does not, for example, comment on the fact that NIPT does detect true cases of genetic disorders, some of which are extremely rare. It also suggests that the only reason NIPT providers expanded their tests was to out-compete one another and make money. That they’re just another flavor of silicon valley biotech looking to make a profit.

The timing of this article is hardly accidental. The verdict of the Theranos trial (Elizabeth Holmes) has a great many people — especially the wealthy investors who lost money — feeling quite uncharitable about technology firms that over-promise on medical testing. This NYT article is written to draw on the parallels — the promises made, the misleading marketing, the billions of dollars to be made. Yet there’s at least one key distinction: NIPT actually works.

Another distinction, and a nuance probably missed by the reporters, is that the customers of the tests are not really the pregnant mothers, but the clinicians who care for them. NIPT is not a direct-to-consumer test like Ancestry or 23andMe. One reason that NIPT and other panel tests continue to expand is because medical professionals want them to. Yes, many of the conditions being tested for are individually quite rare, but collectively they are not. Furthermore, many of these conditions are actionable. Let me take one example that the reporters enjoyed highlighting in their colorful circle plot graphics.

1p36 deletion syndrome, also called monosomy 1p36, is the most commonly observed terminal deletion in the human population, with an estimated prevalence of 1 in 5,000. This syndromic condition comprises many common clinical features including growth deficiency, brain malformations, seizures, craniofacial dysmorphism, congenital heart defects, and hearing/vision problems. Almost all patients have congenital hypotonia (muscle weakness), which is associated with feeding difficulties and developmental delays. Intellectual disability is also common, but variable in severity.

Most cases of 1p36 deletion are sporadic, meaning there’s no family history. Most patients survive well into adult life, but the severity of the disease and its ultimate effects varies widely. According to the Orphanet page for this condition:

Management should be multi-disciplinary and include a regular follow-up. Early diagnosis and access to personalized rehabilitation therapies focusing on motor development, cognition, communication, and social skills are highly recommended.

This is a severe disease with lifelong medical issues, but many of them can be managed. Congenital heart defects may require surgery. Seizures can be treated with standard anti-epileptic medications. Infantile spasms are responsive to corticotrophin. Feeding and growth should be monitored, especially early in life.

There were 3.6 million babies born last year in the US alone; based on the prevalence estimate, 720 of them have 1p36 deletion. Yes, based on the false positive rate, another 4,500 could receive a false diagnosis by NIPT, but the syndrome would be clinically obvious well before birth. And there are still 720 babies who would be correctly diagnosed. Given all of the potential benefits of early/multidisciplinary intervention, it seems like a non-invasive screening test is still a good idea.

This year marks the 20th anniversary of the publication of the human genome reference sequence. As I enjoy recounting to people outside of the genomics field, the investment required to complete that initial assembly is staggering: ten years, dozens of laboratories, hundreds of sequencing instruments, and a billion dollars. Today, using the latest next-generation sequencing, we can sequence a human genome in about two days for a few thousand dollars (yes, a “$1000 genome” is feasible, but only in terms of reagent costs and at centers that can sequence at factory scale).

The human genome reference, advances in sequencing technology, and many years of prolific disease gene discovery have facilitated the widespread adoption of genetic testing as a frontline diagnostic tool. Most single-gene, gene panel, exome, and whole-genome tests now use next-generation sequencing. They have another commonality as well: most rely on alignment of those sequencing reads to “build 37” of the human genome sequence, which dates back to 2003.

There is a newer, better assembly of the human genome: build 38, which has been available for…. (checks watch)… about eight years. Build 38 offers several key advantages over its predecessor, as highlighted in 2017 by the Genome Reference Consortium:

1. Resolution of assembly errors and gaps associated with complex haplotypes and segmental duplications
2. Base-pair–level updates for sequencing errors
3. Addition of “missing” sequences, with an emphasis on paralogous sequences and population variation
4. Better sequence representation for certain difficult genomic structures, such as centromeres and telomeres.

However, Build 38 also comes with a significant cost: it changes the coordinates of genomic loci. In other words, a SNP’s position on build 37 is different (most of the time) on build 38. The same is true of genes and other annotations of the genome: everything has to be re-mapped on Build 38.

LiftOver versus Remapping to Build 38

The UCSC Genome Browser has a useful tool, called liftOver, which allows one to convert coordinates between different versions of the reference assembly. You provide a BED file, select the genome assemblies to convert to/from, and it will produce an output BED file with coordinates based on the desired assembly. Liftover works *pretty* well when one is attempting to obtain the coordinates for regions that are accurately represented in both genome assemblies, i.e. the only thing that has changed is the position number. That’s the case for ~95% of things one might need to convert. The problem is addressing the other 5% of loci that don’t have 1-to-1 unique map locations between assemblies.

For a possible liftOver use case, consider the gnomAD database — which contains variant allele frequencies from large human study cohorts (~120,000 exome sequences and ~18,000 genome sequences). Obviously, gnomAD and its exon-focused predecessor (ExAC) have proven exceptionally valuable when interpreting genetic variants because they offer a somewhat accurate estimate of worldwide prevalence [in the populations that are represented, at least]. Not all of the individuals in gnomAD are perfectly healthy — after all, many of them were part of large cohort studies of common complex disease — but the curators excluded individuals with severe congenital disorders. Thus, an X-linked variant that’s hemizygous in dozens of males in gnomAD is unlikely to cause a rare X-linked recessive disorder.

As noted in the flagship gnomAD paper, the initial WGS dataset yielded 14.9 million high-quality variants from the WES dataset and 229.9 million variants from the WGS dataset, all on build 37. A liftOver to build 38 is available for download. However, if 95% of variants were successfully converted to GRCh38 coordinates using liftOver, we’re still missing 750,000 exome variants and 11.5 million genome variants. These variants, when they’re identified in patients and cohorts with sequence aligned directly to build 38, will [incorrectly] appear novel to gnomAD and thus expected to be quite rare in human populations. The result is a much-diminished signal-to-noise ratio for extremely rare variants in analyses that rely on liftOver data.

Key Genomic Resources Moving to Build 38

Many of the key resources for genome analysis and variant interpretation have fully embraced GRCh38. The ClinVar database and the UCSC Genome Browser, for example, now default to build 38 coordinate systems while providing backward compatibility with build 37. NCBI’s dbSNP database moved to build GRCh38/hg19 with release 143 in March 2015. Many commercial tools are fully compatible with both genome assemblies, including VarSome, which we use in-house.

Other resources are increasingly supporting build38 analysis but still in a transition period. Although the UCSC Genome Browser database defaults to GRCh38, anyone familiar with its annotation tracks will notice that many of them are still missing from the newer assembly. This is an unfortunate reality of having infrastructure entrenched in a certain genome version. The best option would be to ask the groups who contributed key annotation tracks to re-generate their datasets on the new genome assembly. That’s a big ask, especially for groups like the ENCODE Project who have massive datasets on build 37 coordinates.

The vital gnomAD database, unfortunately, falls into the partial-transition category. The liftOver versions of gnomAD have been available for some time, and the latest release of gnomAD included a larger WGS dataset all mapped to GRCh38 (it has other advantages too, like better representation of minority populations). However, the critical WES dataset of gnomAD has yet to be re-mapped, and the gnomAD team recommends sticking with the old version for analysis of coding regions. This is a problem, since that’s where most clinical variant interpretation happens, and undoubtedly contributes to why many clinical labs are reluctant to make the switch.

The good news is that gnomAD plans to make another release, with WES data fully remapped to GRCh38, sometime in 2021. We can expect that in October, since the gnomAD team likes to make their big announcements around the time of the annual ASHG meeting.

Let’s Move to Build 38. Now, and Together

I firmly believe it’s time for the human genetics community as a whole to make a concerted effort to move to build 38 in 2021. In fact, I challenge everyone to do so. The more stakeholders — research groups, clinical laboratories, databases, and even journals — who embrace build38, and consider it standard, the better this will be. Yes, there are bound to be hiccups. Having gone through this transition before, I’d like to offer a few bits of advice:

• Before, during, and after the transition, train yourself to indicate the assembly version whenever providing chromosomal positions or coordinates (in e-mails, Excel files, presentations, etc). Make it a habit.
• Establish shared resources for converting between coordinate systems — such as a local installation of the UCSC liftOver tool — and use a common nomenclature for files/folders that makes the genome version obvious.
• Plan to conduct analyses on both build 37 and build 38 during the transition period, and systematically compare results to make sure that everything is working as it should be.

There are other more subtle differences that we’re likely to encounter as we move analyses to the new genome assembly — such as differences in read mapping performance given the better representation of duplicated sequences and alternate haplotypes. GRCh38 will have its quirks, and we’ll be better off if we muddle through them together.