Inclusive ancestry group assignment in genetics research

Figure 1 shows a breakdown of participants from the 100,000 Genomes Project that could be confidently assigned** to GDAGs. They are labelled according to one of five super populations, using the 1000G as a reference cohort.

As expected from this UK-based project, EUR assigned participants make up around 82% of sequenced participants, with members of other assigned groups ranging from 9% (SAS) to 0.04% (AMR) of the dataset.

Looking closer, however, we see that 6% (n = 4,498) of participants from the 100,000 Genomes Project remain unassigned using this approach. That is, they don’t appear ‘genetically similar’ enough to any of the five super populations, and so cannot be confidently labelled.

Unfortunately, this means they are more likely to be excluded from downstream research they have consented to be involved in – including research into the 100,000 Genomes Project dataset.

Concerns surrounding not fully using data from diverse and under-represented populations are well documented. They have led to the systematic exclusion of many participants from analyses of large genetic datasets, such as the UK Biobank (where the 1000G five continental super population model was typically employed).

Group assignment using the UK Biobank as a reference cohort

There are multiple reasons why individuals can end up not being assigned to a GDAG.

The first is inherent to all reference-based group assignment approaches. That is, individuals with genetic ancestors best represented by multiple reference populations, will not be confidently assigned in many cases. For example, somebody with a European mother and an East Asian father.

Even in instances where assigning individuals to single, discrete GDAGs is still useful or necessary, another practical limitation remains: the availability and selection of reference populations.

While widely used, the 1000G is not representative of global genetic diversity. This means that participants from the 100,000 Genomes Project with recent genetic ancestors that fall outside of these five broad, continental regions, are also unlikely to be confidently assigned using the 1000G as a reference cohort.

This issue of representativeness can be demonstrated using an alternative reference cohort. Using data from around half a million people in the UK Biobank, Florian Privé described an expanded set of reference populations defined by an individual’s country-of-birth, self-reported ethnicity, and crucially – their genetic similarity to one another.

Figure 2 shows GDAG assignments from the 100,000 Genomes Project, generated using the expanded set of reference populations from the UK Biobank.

Using this updated approach, 1,773 individuals in the 100,000 Genomes Project that were previously unassigned using the 1000G (Figure 2), can now be assigned using the same stringency threshold.

Notably, 72% of these newly assigned individuals are assigned to either the East Africa (EAF), Middle East (MID), or North Africa (NAF) groups. All of which have no close genetic analogue among the 1000G reference populations.

New insights gained from a more inclusive group assignment approach

We can use an example from my own work on rare condition diagnosis to demonstrate some advantages of taking a more inclusive approach to describing genetic diversity.

As part of my analysis, I looked for differences in the number of prioritised (or “tiered”) candidate pathogenic variants in probands with rare conditions, each of whom were assigned to a super population in the 1000G. I found that EUR assigned probands often had fewer reported candidate variants than those assigned any other group.

I then discussed how variant filtering based on observed allele frequencies from imbalanced genomic datasets can contribute to any differences.

In Figure 3, I replicated this analysis using GDAG assignments. This was based on a participant’s genetic similarity to the expanded selection of reference populations generated from the UK Biobank.

Using these updated GDAGs, several new insights emerge. For example, we can now see that probands in the newly assigned EAF group have (on average) considerably more candidate variants within the applied PanelApp gene panel. They also reported to clinicians (Tier 1 or 2) an average of 2.9 times more than EUR assigned probands, and 1.8 times more than AFR assigned probands (Figure 3d/e).

A similar pattern is observed among Tier 3 candidate variants that are not observed within the applied PanelApp gene panel (Figure 3b/c). This suggests that observations are due to genetic divergence of EAF assigned probands from populations used for allele frequency estimation, and not, for example, gene panel ascertainment.

This makes sense, as estimates have indicated that lineages ancestral to East African and West African populations separated tens of thousands of years ago.

Many East African populations are therefore highly differentiated from those present in the African (AFR) reference datasets used for allele frequency filtration. Such reference datasets are instead made up almost entirely of West Africans and African Americans (see gnomAD v3.1).

As a result, variants common specifically in East African populations will appear to be superficially rare. Probands genetically similar to these East African populations will have an increased number of variants falsely reported as good pathogenic candidates.

Indeed, these results highlight the continued need for under-represented communities to be sequenced and included in available allele frequency catalogues. They suggest that East African communities should be considered as a priority for future research and engagement.

Final thoughts

These insights show that a more inclusive approach to describing the genetic diversity of 100,000 Genomes Project participants has the potential to inform clinical practice in diagnosing rare conditions.

This is particularly significant for individuals from historically under-researched groups.

The blog here demonstrates this by simply increasing diversity in reference populations to assign individuals to GDAGs (Figure 2). By doing this, we could identify participants genetically similar to a greater diversity of worldwide populations, enabling more participants to be included in downstream research.

Researchers can use this information in their own work by accessing the National Genomic Research Library (NGRL). Here, all of the GDAG assignments (and accompanying statistics) generated using the expanded UK Biobank reference cohort (Figure 2/3) have been made available for all participants.

Importantly, while use of this expanded reference cohort offers a simple, accessible means of improving GDAG assignment, even this approach is insufficient. To truly capture the underlying complexities of human genetic ancestry, we often need to go beyond broad continental groupings.

This could allow us to describe genetic structure at a finer scale, or indeed, challenge the need for single discrete groups as reliable proxies for genetic ancestry in the first place.

To read wider discussions surrounding this topic, see the Reading List below. You can also read about other research in the field of genomics by checking out our bioinformatics or research blogs.

Data availability

Documentation on how to use and access updated GDAG information for all participants in the NGRL for researchers with access to the Genomics England Research Environment can be found in our research documents area.

Footnotes

* We use the term “genetically defined ancestry group (GDAG)” here, whilst noting that assigning individuals to groups in the manner described above is simply a measure of their genetic similarity to a given reference group. Such group labels do not precisely nor completely describe a participant’s genetic ancestry.

** Confidence as used here refers the maximum allowable genetic distance between a target individual and a given reference population to be assigned as such. In this blog post - as in Prive et al. 2022 - Euclidean distance on PC space between the reference population and the target individual is converted into an approximate FST, and a threshold of FST<0.002 is utilised as the minimum requirement to assign an individual to a given GDAG. The number of individuals assigned to a GDAG therefore has the potential to increase when a more lenient threshold is applied.

Reading list

Graham Coop on “ancestry groups” vs genetic similarity

Getting ancestry right for science and society

What is ancestry?