Scientists Build a Genetic Score for Alzheimer’s Disease that Works Across Ancestries

By Dr. Chinta SidharthanReviewed by Laura ThomsonSep 7 2026

Most polygenic risk scores for Alzheimer's disease have been built almost entirely from European genetic data, which has substantially limited their usefulness in other populations. In a recent study published in Nature Genetics, a team led by researchers at Boston University developed and tested a new risk score designed to work across several ancestry groups. They evaluated its accuracy against clinical diagnosis, brain biomarkers, and cognitive performance across participants from different ancestries.

Alzheimer's Background

Alzheimer's disease is a primary cause of dementia among older adults, resulting in declining daily functioning that can unfold over the lifetime of the patient.

A large body of research has shown that genetic factors shape the likelihood of developing the disease and the pace of its progression. Moreover, while the apolipoprotein E (APOE) ε4 allele is the strongest known genetic risk factor, genome-wide association studies have also identified numerous additional variants that each contribute a small amount to overall risk.

Polygenic risk scores combine these variants into a single measure of inherited susceptibility. Most of these scores, however, come from datasets dominated by people of European ancestry, and their accuracy drops noticeably when applied to other populations. However, expanding genetic datasets from African American, Caribbean Hispanic, and East Asian cohorts now offer a chance to build risk scores that can be applied to a wider range of people.

The Study

The researchers looked at whole-genome sequencing and phenotypic data from 36,362 participants in the Alzheimer's Disease Sequencing Project, which spanned six ancestry clusters, including European, African American, Caribbean Hispanic, Native American Hispanic, South Asian, and East Asian groups.

Genome-wide association summary statistics came from several independent sources, including the European Alzheimer's Disease Biobank, FinnGen, the Million Veteran Program, and Japanese and Korean genetic studies, with the smaller East Asian datasets combined through meta-analysis.

To build the risk score, the team applied several statistical approaches, including various clumping and thresholding methods. The team tested three strategies for combining results across ancestries, including a meta-analysis of all datasets, an unweighted summation of separately constructed scores, and a weighted summation using coefficients derived from an independent genotype and phenotype resource, the Alzheimer's Disease Genetics Consortium. The authors evaluated each candidate score for its ability to predict Alzheimer's diagnosis within the sequencing project cohort.

Because the APOE region carries an outsized genetic effect, the researchers recalculated the best-performing score after removing variants within one megabase of the APOE gene, and then modeled APOE genotype as a separate factor. They tested this adjusted score in four additional groups:

• A South Korean cohort
• A large multi-ancestry sample from the All of Us Research Program
• An African American cohort from Rush University Medical Center
• Participants of European ancestry from the Framingham Heart Study

Lastly, the team examined how the score related to several disease markers beyond diagnosis itself. These included amyloid and tau levels measured in cerebrospinal fluid and blood plasma, postmortem measures of plaque and tangle burden, longitudinal scores for memory, executive function and language, and structural brain scans that measured hippocampal volume.

Key Findings

The study showed that the multi-ancestry genetic risk score was strongly associated with Alzheimer's diagnosis, with each standard deviation increase corresponding to roughly a 39% increase in the odds of Alzheimer’s diagnosis. The association was strongest in individuals of European ancestry and progressively weaker in the Caribbean Hispanic, African American, and Native American Hispanic groups.

While the researchers found no significant association in South Asian or East Asian groups, they attributed this to small sample sizes rather than a lack of genetic relevance. The findings held up across four validation cohorts, including the South Korean sample, the All of Us large multi-ancestry group, the African American cohort, and the European ancestry cohort.

Furthermore, higher scores were consistent with biological markers of disease, including lower amyloid-beta and higher tau levels in cerebrospinal fluid, and a greater likelihood of biomarker profiles consistent with active amyloid and tau pathology. Similar patterns also appeared in blood-based tau measurements, with associations often stronger in women than men. Higher scores were also linked to more severe postmortem plaque density and tangle spread, which corroborates established staging systems.

On the cognitive testing front, higher scores corresponded to weaker memory, language, and executive function, with the gap between high- and low-scoring groups widening with age. This relationship was most pronounced before formal diagnosis, which suggested that the score reflects early differences in cognitive reserve rather than the pace of decline after disease onset.

Reduced hippocampal volume also correlated with higher risk scores. However, the researchers noted that continued underrepresentation of non-European groups, such as East and South Asian populations, in both discovery and validation datasets could limit how well some of these findings generalize.

Conclusion

The study introduced a genetic risk score that performs meaningfully across a wider range of ancestries than earlier versions, while still leaving room for improvement among underrepresented groups. Moreover, by linking genetic risk with measurable changes in cognition, brain structure, and biomarker levels, the study provides a foundation for identifying people who may benefit from earlier monitoring, provided future studies expand the genetic data available for populations still poorly represented in research.