Population-specific genetic risk scores advance precision medicine for Han Chinese populations

Researchers at Academia Sinica have developed the first population-specific polygenic risk score (PRS) models for people of Han Chinese ancestry, achieving unprecedented accuracy in predicting risks for common diseases such as diabetes, heart disease, autoimmune disorders. The study, published in Nature on October 15, 2025, analyzed genomic and health data from more than half a million Taiwanese participants of the Taiwan Precision Medicine Initiative (TPMI) and demonstrates the transformative potential of ancestry-specific genetics in improving precision medicine across East Asia.

Unlocking population diversity in genomic research

Polygenic risk scores combine the effects of millions of genetic variants to estimate an individual’s likelihood of developing complex diseases. However, most existing PRS models were built from studies of European ancestry populations, leading to reduced accuracy when applied to others.

To close this gap, TPMI scientists conducted the largest genome-wide association study (GWAS) of Han Chinese individuals to date, analyzing 695 diseases and 24 quantitative traits across 463,447 participants. The team identified 2,656 independent genetic associations, including 95 previously unreported variants, and found that the genetic risks collectively explain up to 10.3% of total health variation in the Taiwanese population.

“This project marks a milestone for precision medicine in East Asia,” said Dr. Cathy S.-J. Fann, senior corresponding author at the Institute of Biomedical Sciences, Academia Sinica. “By integrating large-scale genomic and clinical data, we are building predictive models that truly reflect the real genetic architecture of our population.”

Improved disease prediction through population-specific models

Using state-of-the-art algorithms such as LDpred2 and PRSmix+, the researchers developed and validated PRS models for hundreds of diseases, achieving area-under-the-curve (AUC) values exceeding 0.8 for ankylosing spondylitis and around 0.7 for psoriasis, atrial fibrillation, rheumatoid arthritis, and type 2 diabetes. When applied to independent cohorts from the Taiwan Biobank, UK Biobank, and All of Us Project, TPMI-derived PRS consistently outperformed models trained on European data for East Asian populations.

“Our results show how locally developed risk prediction can substantially enhance healthcare precision,” added Dr. Hung-Hsin Chen, the first and corresponding author at the Institute of Biomedical Sciences, Academia Sinica.

Shared genetic architecture across diseases

Analyses of genetic correlations revealed three major clusters of traits—cardiometabolic, autoimmune/infectious, and kidney-related—suggesting shared biological mechanisms. Multitrait PRS models further improved predictive performance, especially for cardiometabolic diseases, where the explained variance increased 1.77-fold.

Novel insights into hepatitis B and immune diseases

Leveraging Taiwan’s unique hepatitis B (HBV) epidemiology, the team also identified 19 new genetic loci associated with HBV infection and found inverse genetic correlations between HBV and autoimmune diseases such as Sjögren’s syndrome, psoriasis, and systemic lupus erythematosus. These results highlight how population-specific datasets can uncover gene–environment interactions invisible in European cohorts, offering new insights into both infection susceptibility and immune regulation.

Toward equitable precision medicine

The researchers emphasize that genetic diversity is essential for fair global health. Despite representing almost a quarter of the world’s population, East Asian people account for under 4% of participants in past genome-wide studies. By making TPMI data available for research, the team hopes to inspire similar efforts worldwide.

“Precision medicine should serve everyone, not just the populations that have been most studied,” said Dr. Chen. “Our work offers a blueprint for other countries to develop population-specific risk prediction frameworks.”

Provided by
Academia Sinica