Human demographic history impacts genetic risk prediction across diverse populations

By Alicia R. Martin, Christopher R. Gignoux, R. Walters, G. Wojcik, Benjamin M. Neale, Simon Gravel, Mark J. Daly, Carlos D. Bustamante, Eimear E. Kenny. 2016

1,291 citations recorded by the source corpus.

The vast majority of genome-wide association studies are performed in Europeans, and their transferability to other populations is dependent on many factors (e.g. linkage disequilibrium, allele frequencies, genetic architecture). As medical genomics studies become increasingly large and diverse, gaining insights into population history and consequently the transferability of disease risk measurement is critical. Here, we disentangle recent population history in the widely-used 1000 Genomes Project reference panel, with an emphasis on populations underrepresented in medical studies. To examine the transferability of single-ancestry GWAS, we used published summary statistics to calculate polygenic risk scores for six well-studied traits and diseases. We identified directional inconsistencies in all scores; for example, height is predicted to decrease with genetic distance from Europeans, despite robust anthropological evidence that West Africans are as tall as Europeans on average. To gain deeper quantitative insights into GWAS transferability, we developed a complex trait coalescent-based simulation framework considering effects of polygenicity, causal allele frequency divergence, and heritability. As expected, correlations between true and inferred risk were typically highest in the population from which summary statistics were derived. We demonstrated that scores inferred from European GWAS were biased by genetic drift in other populations even when choosing the same causal variants, and that biases in any direction were possible and unpredictable. This work cautions that summarizing findings from large-scale GWAS may have limited portability to other populations using standard approaches, and highlights the need for generalized risk prediction methods and the inclusion of more diverse individuals in medical genomics.

Subfields: Genetics, Computational Biology & Bioinformatics

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Genetics

In Computational Biology & Bioinformatics

Citations in the corpus

Underspecification Presents Challenges for Credibility in Modern Machine Learning (2020, background): “, 2018), but face difficulties when applied to new patient populations (Martin et al., 2017; Duncan et al., 2019; Berg et al., 2019).”
PRSice-2: Polygenic Risk Score software for biobank-scale data (2019, neutral): “For PRS analyses performed in family data or across diverse populations, for instance, results should be interpreted carefully [14] and extensions of the standard PRS approach or alternatives may be required [14,28–30] to generate more informative results.”
Geographic variation and bias in polygenic scores of complex diseases and traits in Finland (2018, background): “varying linkage-disequilibrium patterns and allele frequency differences, between the target sample and the GWAS data can limit generalizability across populations (Martin et al. 2017; Reisberg et al. 2017).”
Comparative genetic architectures of schizophrenia in East Asian and European populations (2018, neutral): “Importantly, the applicability of training data from EUR studies to those of non-European ancestry has not been fully assessed, leaving uncertainty as to the biological relevance of discoveries made in EUR samples for non-Europeans 21 .”
Tread Lightly Interpreting Polygenic Tests of Selection (2018, background): “These initial approaches focused on comparison across populations, but comparisons can also be made across time within single populations using either ancient DNA (Mathieson et al. 2015) or using haplotypic signatures of allele frequency change (Field et al. 2016).”
Integrative omics for health and disease (2018, background): “A major obstacle remains in building such frameworks for multiple omics profiles, which is likely to face some of the same challenges, such as the difficulty in applying results discovered in one population to individuals in another 79,80 .”
Adaptive Sampling Strategies to Construct Equitable Training Datasets (2022, background): “For instance, a deep learning approach to training PRSs might use all available data for a single model instead of training separate models such as in our example [5].”
Polygenic risk score for schizophrenia is more strongly associated with ancestry than with schizophrenia (2018, background): “Other studies have also shown that the PRS for schizophrenia varies between populations (Martin et al., 2017).”