Bridging genomics' greatest challenge: The diversity gap

Abstract

Achieving diverse representation in biomedical data is critical for healthcare equity. Failure to do so perpetuates health disparities and exacerbates biases that may harm patients with underrepresented ancestral backgrounds. We present a quantitative assessment of representation in datasets used across human genomics, including genome-wide association studies (GWASs), pharmacogenomics, clinical trials, and direct-to-consumer (DTC) genetic testing. We suggest that relative proportions of ancestries represented in datasets, compared to the global census population, provide insufficient representation of global ancestral genetic diversity. Some populations have greater proportional representation in data relative to their population size and the genomic diversity present in their ancestral haplotypes. As insights from genomics become increasingly integrated into evidence-based medicine, strategic inclusion and effective mechanisms to ensure representation of global genomic diversity in datasets are imperative.

Keywords: GWAS; ancestry; bias; direct-to-consumer; diversity; equity; genomics; global datasets; healthcare; inclusion; pharmacogenetics; representation.

Graphical abstract

Figure 1

Sampling bias Left, number of genome-wide association study (GWAS) participants of European ancestry in 2009 from the GWAS Catalog. Right, update by Popejoy and Fullerton (2016) on the ancestry breakdown in GWASs in 2016. Figure adapted from Popejoy and Fullerton.

Figure 2

GWAS Diversity Monitor Left, number of GWAS participants by ancestry, including different types of GWASs or health conditions (parent terms), discovery stages, 2023. Right, number of GWAS participants across all parent terms, discovery stage, 2024. Accessed online 9 Sept 2024. We note inconsistencies in labeling and coloring of populations between the figures due to different ways of reporting ancestries by sources.

Figure 3

Total proportion of participants in the GWAS Diversity Monitor The total proportion of representation from any population other than Asian (3.96%) or European (94.48%) is <1%, suggesting that current efforts toward diversity in genomics are failing. Source: GWAS Diversity Monitor (https://gwasdiversitymonitor.com/).

Figure 4

Biased representation and missing global diversity in pharmacogenomics (A) Pie charts reflect the percentage of individuals included in PharmGKB-curated studies with respect to the total number of individuals. Europeans (EUR) make up 63.6%, 28.1% east Asian (EAS), 2.2% central/south Asian (SAS), 2.1% African American or Afro-Caribbean (AAC), 1.6% Sub-Saharan African (SSA), 1.6% Latino (LAT), 0.9% Near Eastern (NEA), 0.1% Indigenous American (AME), and 0% Oceanian (OCE). (B) Difference in percentage of ancestries between global census and representation in pharmacogenetic studies. A percentage of 0 represents a balanced proportion as compared to the share of the population globally. We note inconsistencies in labeling and coloring of populations due to different ways of reporting ancestries by sources. (Rough estimates of global biogeographical populations, including their diaspora, were calculated using sources available in Table S1.)

Figure 5

Individuals taking part in clinical trials between 2015 and 2019 segmented by population categories reported by the FDA “Other” includes populations such as Latino or Oceanians, whose lack of data is particularly evident. We note inconsistencies in labeling and coloring of populations due to different ways of reporting ancestries by sources. Data adapted from the FDA drug trial snapshot.

Figure 6

PharmGKB predicted metabolizer phenotype frequencies for CYP2D6, according to biogeographical groupings used in the resource These data were adapted to reflect definitions of allele activity, where ultrarapid metabolizer has an activity score >2.25, normal metabolizer 1.25–2.25, intermediate metabolizer 0.25–1, poor metabolizer 0, and indeterminant metabolizer not applicable.