DNA Alone Isn’t Enough: Why the Future of Genomics Is About Integration

Genomelink

Introduction

This week, I came across three new genomics papers that made me pause. They covered heart failure, polygenic risk scores, and the integration of electronic health records. All three pointed to the same message: DNA alone isn’t enough.

As someone who has worked in consumer genomics since 2018, that conclusion feels both obvious and revolutionary. Obvious, because health has always been multi-factorial. Revolutionary, because we are finally starting to see large-scale evidence that combining different data layers actually works.

My Entry into Consumer Genomics

I’ve been reading GWAS papers since 2018, back when my job was annotating SNP-level data — pulling out variants tied to individual traits.

What struck me then was this: while entertainment-oriented traits like personality, nutrition, or exercise performance were rarely studied, the same SNPs often showed up in research on cardiovascular disease, diabetes, or psychiatric conditions. That overlap fascinated me — and it was exactly what made consumer genomics exciting at Genomelink.

Consumer genomics has always been a turbulent industry.

• 2007–2010: The dawn of direct-to-consumer DNA kits. 23andMe and AncestryDNA launched, but adoption was limited and many early companies folded.
• 2015: The boom began. 23andMe passed one million customers.
• 2018: The holiday craze. Kits became Christmas gifts in the US, and the global market surged to ~25 million users.
• 2019–2020: The slowdown. Growth plateaued; industry leaders revised their outlooks.
• 2023–2025: Recovery. Estimates suggest 38–50 million users worldwide, growing steadily but without the explosive pace of the past.

Having lived through this cycle, I know how fragile hype can be. That’s why breakthroughs like the three studies below matter: they show how genomics can move beyond the early hype toward lasting, clinical relevance.

1. Heart Failure: Rare + Common Variants

A large-scale study led by the University of Pennsylvania analyzed nearly 2.4 million individuals, identifying 176 common variants associated with heart failure, including 105 never reported before.

They also confirmed the role of rare variants in genes such as TTN, MYBPC3, FLNC, and BAG3, underscoring that rare and common variants together shape disease risk. Among these, TTN and MYBPC3 are particularly well-known in cardiomyopathy and heart failure research, often cited as “classic” examples when discussing genetic contributions to the disease. (1)

Michael Levin (University of Pennsylvania) explained:

“Aggregating the effects of common genetic variants, in the form of a polygenic score, we can begin to better explain genetic risk of heart failure.”

This is a reminder that risk is not binary — it’s layered, and the full picture emerges only when we consider both rare and polygenic contributions.

2. OmniPRS: Functional Annotations Boost Accuracy

A team from Huazhong University of Science and Technology introduced OmniPRS, a scalable framework that integrates functional annotation data — both tissue-specific and non-tissue-specific — into polygenic risk score construction. Unlike conventional PRS methods that rely mainly on GWAS summary statistics, OmniPRS quantifies variance across multiple annotation categories and combines them into an aggregated score.

Results were impressive across a broad range of traits. In 135 simulations and validation using UK Biobank and 1,000 Genomes Project data, OmniPRS consistently outperformed existing approaches:

For quantitative traits such as height, BMI, and lipid levels (LDL, HDL, triglycerides, total cholesterol), prediction accuracy improved by an average of 52.3% compared to the traditional “clumping and thresholding” method, and by 8.4% compared to a Bayesian framework.

For binary traits such as Alzheimer’s disease, schizophrenia, bipolar disorder, hypertension, and type 2 diabetes, OmniPRS also showed clear improvements in predictive performance relative to current state-of-the-art approaches. (2)

Xingjie Hao (Huazhong University of Science and Technology) emphasized:

“Leveraging multiple functional annotation categories could significantly improve the accuracy of polygenic prediction.”

Takeaway: PRS is evolving from a statistical construct into a biologically-informed model that captures functional context, opening new possibilities for clinical translation.

3. EHR × PRS: Complementary Information

An international team (Helsinki, Broad, MGH) analyzed health data from 845,929 individuals in FinnGen, UK Biobank, and Estonian Biobank.

They focused on 13 common conditions, including type 2 diabetes, coronary heart disease, atrial fibrillation, asthma, major depressive disorder, knee/hip osteoarthritis, gout, and several cancers (lung, colon, breast, prostate).

The researchers showed that phenotype risk scores (PheRS) derived from electronic health records captured independent information from PRS across these diseases. When integrated, the two approaches improved predictive performance compared to either alone. (3)

Andrea Ganna (University of Helsinki’s Institute for Molecular Medicine, the Broad Institute, and Massachusetts General Hospital) summarized:

“Combining EHR and genetic data can be an advantageous strategy for the prediction of many common diseases.”

In other words: DNA and clinical records are complementary. Integrated, they produce risk models that are more practical and clinically relevant.

4. Shared Message: Integration Is the Future

Different methods, same conclusion:

Rare + Common variants (heart failure study) → disease risk isn’t driven by rare mutations alone, but also by the cumulative effects of many common variants.

PRS + Functional annotations (OmniPRS) → adding biological context makes genetic predictions far more accurate.

PRS + EHR data (PheRS study) → medical history provides complementary information that DNA alone can’t capture.

Put together, these findings highlight a simple truth: no single data source is enough to explain complex diseases. The real progress comes when we combine multiple perspectives — genetic, biological, and clinical — into one integrated model.

As Levin put it:

“We may need to consider a broader spectrum of genetic variation … when we think about clinical risk.”

In other words, while Levin was referring to common and rare variants, the broader lesson across all three studies is clear: the future of disease prediction isn’t about choosing between rare vs. common variants, or DNA vs. health records. It’s about integration across all layers of data to give a fuller and more actionable picture.

Conclusion: From Ancestry to Health

YourRoots

For me, these studies reinforce something I felt back in 2018: genomics is not about isolated SNPs, but about connections.

That’s also why I see potential in ancestry and family history platforms like YourRoots. Ancestry data is often the first step people take into understanding themselves. One day, it may connect seamlessly with genomic and health data, letting people see both their past and their future in a single integrated experience.

I’ve been lucky to see genomics from many angles: selected as a postdoc at a national genomics institute, part of the founding team at Genomelink, and was mentoring overseas researchers at Beyond Next Ventures.

I’m sharing these thoughts because I believe genomics belongs not just in labs, but in people’s lives.

Akiko Yoshida

Portfolio->https://lumpy-lyre-629.notion.site/Akiko-Yoshida-Portfolio-221e2fffd7e380cea33af4a1d630ab5e?pvs=143