A 565,390-person Taiwanese cohort tests polygenic scores outside European ancestry

Han Chinese people comprise nearly 20 percent of the global population but remain under-represented in genetic studies [s1]. Two papers published in Nature on 15 October describe an attempt to change that at scale, and then to test what the resulting data can actually predict.

The cohort

The Taiwan Precision Medicine Initiative, established by Academia Sinica with 16 major medical centres around Taiwan, has recruited 565,390 participants who consented to provide DNA for genetic profiling and to grant access to their electronic medical records for research [s1].

Two design choices distinguish it from a conventional biobank. Medical record access is both retrospective and prospective, which allows longitudinal study rather than a single cross-sectional snapshot [s1]. And genotyping uses arrays of single-nucleotide polymorphisms optimised for people of Han Chinese ancestry rather than arrays designed around European variation [s1].

That second point is more consequential than it sounds. Genotyping arrays capture common variants efficiently by exploiting patterns of correlation between nearby sites, and those patterns differ between ancestries. An array designed for one population is a blunter instrument in another.

Participants also agreed to be re-contacted for future research and to receive personalised genetic risk profiles with health management recommendations [s1]. The initiative established a data access platform intended to secure the data while facilitating academic research [s1].

What the scores do

The companion paper used the cohort's phenotypic data to run genomic analyses across the medical phenome in individuals genetically similar to Han Chinese reference populations, identifying population-specific genetic risk variants and new findings for various complex traits [s2].

From those, the researchers developed polygenic risk scores — composite estimates that sum the small effects of many variants into a single risk figure — and report strong predictive performance for cardiometabolic diseases, autoimmune disorders, cancers and infectious diseases [s2].

The headline quantity is that the identified genetic risks accounted for up to 10.3 percent of the overall health variation in the cohort [s2].

Reading 10.3 percent correctly

That figure is an upper bound across the traits examined, not a typical value, and "health variation" is a population-level statistic rather than a statement about any individual.

A score explaining 10 percent of variation in a trait can still be useful — it can stratify a population into risk tiers well enough to change screening intervals — while telling a given person very little about whether they personally will develop a condition. The gap between population discrimination and individual prediction is the most persistent source of confusion about polygenic scores, and it does not close by adding participants.

The replication, and its limits

The authors report consistent findings in an independent dataset, Taiwan Biobank, and among people of East Asian ancestry in the UK Biobank and the All of Us Project [s2].

Replication in an independent Taiwanese cohort is meaningful. Replication in East Asian-ancestry participants within UK Biobank and All of Us is a weaker test than it appears — those subgroups are comparatively small within their parent cohorts, and East Asian ancestry is not a single genetic category. Transferability from a Taiwanese-derived score to, say, a Japanese or Vietnamese population is a different question and one these data cannot answer.

Both papers are careful about the ancestry framing, describing the analysis population as individuals genetically similar to Han Chinese reference populations [s2] rather than treating ancestry as a self-evident label.

Why non-European cohorts matter for everyone

Most genetic studies have focused on populations of European ancestry, creating a global imbalance in precision medicine and underscoring the need for research in non-European groups [s2]. The practical consequence is that polygenic scores derived from European cohorts predict poorly in others — not because the underlying biology differs fundamentally, but because the statistical machinery depends on correlations between measured markers and unmeasured causal variants, and those correlations are ancestry-specific.

Building a score from within a population is the only reliable fix. That is what makes a half-million-person Han Chinese cohort with linked medical records a resource rather than a duplication of existing work.

What has not been demonstrated

Neither paper reports a clinical trial. The first frames the cohort's potential uses as validating genetic risk prediction models, performing clinical trials of risk-based health management, and informing health policies [s1] — all prospective.

Whether returning personalised genetic risk profiles to participants improves their health is untested here. Whether a polygenic score changes a screening decision in a way that produces better outcomes than existing clinical risk factors is untested here. Those are the questions that determine whether polygenic risk scores become clinical tools or remain research instruments, and a large well-genotyped cohort is a precondition for answering them rather than an answer.

What to watch

The initiative's own stated ambition is to serve as a model for population-based precision medicine in other diverse populations [s1]. The test of that is whether the risk-based health management trials the cohort was built to enable actually get run, and what they show. Recruiting 565,390 people is the easier half.

Sources

• The Taiwan Precision Medicine Initiative provides a cohort for large-scale studies — Nature, 15 October 2025
• Population-specific polygenic risk scores for people of Han Chinese ancestry — Nature, 15 October 2025