When polygenic risk scores (PRS) calculated from European-centric studies are applied to diverse global populations, their predictive power frequently degrades. As noted in On cross-ancestry cancer polygenic risk scores by Xiang Zhou, these scores often show significantly lower performance in individuals of African ancestry.
This drop in accuracy is not usually because the fundamental biology of a disease changes from one person to another. Instead, it is a consequence of differences in genetic architecture—the specific frequencies, genomic contexts, and historical patterns of inheritance that shape how genetic variations manifest across human populations. To build clinical models that work for everyone, we have to understand the mechanics behind why genetic risk expressions diverge.
Shifting Haplotypes and Linkage Disequilibrium
In genome-wide association studies (GWAS), researchers rarely discover the exact, physical mutation that causes a disease. Instead, they find "tag" single-nucleotide polymorphisms (SNPs)—variants that happen to be located near the true causal mutation and are inherited alongside it. These blocks of co-inherited variants are known as haplotypes, and the statistical correlation between them is called Linkage Disequilibrium (LD).
Crucially, LD structures are not uniform across the globe. Because modern humans originated in Africa, African genomes have had more time to undergo historical recombination, resulting in shorter, more fragmented LD blocks compared to European or Asian populations.
As a result, a tag SNP that serves as a reliable proxy for a disease variant in one population may be completely uncoupled from it in another. As demonstrated by Seyedeh M. Zekavat et al. in Deep coverage whole genome sequences and plasma lipoprotein(a) in individuals of European and African ancestries, while the primary genetic drivers of a trait are often shared across ancestries, the specific variants and their observed effect sizes diverge due to these ancestral haplotype differences. When the proxy breaks, the predictive power of the genetic test breaks with it.
European Population (Longer LD Blocks)
[ Tag SNP ]---------------------[ Causal Variant ] <- Strongly Correlated
African Population (Shorter LD Blocks due to historical recombination)
[ Tag SNP ]-------//-------[ Recombination ]-------//-------[ Causal Variant ] <- Uncorrelated
Chromosomal Context and Local Ancestry
It is tempting to view a genetic variant's effect size as an intrinsic property of the mutation itself. However, genes operate within an interconnected cellular ecosystem. The risk conferred by a specific genetic variant can be heavily modified by the ancestral background of the exact chromosomal segment where it resides.
This phenomenon is known as local ancestry risk modification. In a study of multi-ethnic cohorts by Ashley Beecham et al. titled Ancestral risk modification for multiple sclerosis susceptibility detected across the Major Histocompatibility Complex in a multi-ethnic population, the authors found that the exact same risk allele could carry twice the statistical weight depending on whether it was inherited on a European or an African ancestral chromosome.
This happens because of epistatic interactions: background genetic variations elsewhere on the chromosome can act as silencers or amplifiers for the risk allele. Failing to account for this chromosomal context leads to inaccurate risk estimates in admixed populations.
Independent Signals and Population-Specific Loci
Beyond shifting contexts for shared variants, distinct demographic histories and evolutionary pressures mean that some populations possess entirely unique genetic risk factors.
For example, by focusing specifically on underrepresented cohorts, Chani J. Hodonsky et al. identified an Amerindian-specific regulatory signal influencing red blood cell traits in Genome-wide association study of red blood cell traits in Hispanics/Latinos: The Hispanic Community Health Study/Study of Latinos. This protective variant simply does not appear in European-centric datasets.
Even when a disease gene is universally relevant across all humans, the precise causal mutations within that gene can differ by population. In Meta-analysis of lipid-traits in Hispanics identifies novel loci, population-specific effects and tissue-specific enrichment of eQTLs, Jennifer E. Below et al. showed that complex traits often involve independent signals across distinct ancestries, even when the broader biological locus is conserved.
The Path to Portable Genetics
If genetic effects were uniform across the genome, cross-ancestry prediction would be simple. However, as Jiacheng Miao et al. demonstrate in Quantifying portable genetic effects and improving cross-ancestry genetic prediction with GWAS summary statistics, genetic covariance is highly localized; a tiny fraction of the genome may account for the majority of shared cross-ancestry correlation, while other regions vary wildly.
To resolve this, researchers cannot simply rely on larger sample sizes from a single demographic. As argued by Esteban J. Parra et al. in Genome-wide association study of type 2 diabetes in a sample from Mexico City and a meta-analysis of a Mexican-American sample from Starr County, Texas, fine-mapping studies within diverse cohorts are required to isolate true causal variants from population-specific tag SNPs. By mapping the unique haplotype structures and local ancestry effects of diverse populations, genomic medicine can move away from fragile proxies and toward universally accurate, biology-driven risk assessment.