HorizonBio

For decades, personalised medicine meant genomics: sequencing the static letters of your DNA and translating variants into risk scores. That was a revolution. But it was only half the story. Your genome is a fixed blueprint; what actually determines your health day-to-day is how that blueprint is read. Enter the epigenome: dynamic, responsive, and far closer to the living experience of disease than any SNP catalogue.

The limits of a static map

A genome-wide association study (GWAS) identifies statistical associations between genetic variants and disease outcomes. The information is real, but it is frozen at conception. It cannot tell you that your metabolic risk escalated this year because you slept poorly, gained visceral fat, or stopped exercising. Your DNA does not change; your biology does.

This is why polygenic risk scores, for all their promise, explain only a fraction of variance in complex diseases. The missing variance lives in the epigenome, in the methyl groups, histone modifications, and non-coding RNA networks that convert static sequence into dynamic physiology.

What epigenomics adds

Unlike genomics, epigenomics gives you a time-stamped readout. A methylation array run today captures the cumulative biological effect of everything your body has experienced: diet, stress, sleep, exercise, infection, and ageing. Interpreted correctly, it is less like a birth certificate and more like a health diary written in molecular ink.

Epigenetic clocks (Horvath, GrimAge, PhenoAge) predict biological age and mortality more accurately than chronological age.^1,2,3
Tissue-of-origin methylation signatures underpin emerging liquid-biopsy approaches for multi-cancer early detection. These technologies are real and advancing, but sensitivity for genuinely early-stage cancers remains modest and is still actively debated - they are not yet an established replacement for imaging or standard screening.
Prospective cohort studies have linked blood methylation at metabolic loci (including ABCG1 and TXNIP) to incident type 2 diabetes years before clinical diagnosis - for example over a mean 8.5-year follow-up.⁴ That is research-grade risk stratification, not a guarantee that any consumer panel can forecast onset on a fixed multi-year clock.
Response to dietary interventions can be monitored via methylation shifts over weeks to months in some intervention studies, though effect sizes and loci vary.

The future of personalised medicine is not about reading the instructions more accurately; it is about understanding when and why the cell chooses to follow them.

From population science to the individual patient

The challenge is translation. Methylation arrays generate hundreds of thousands of data points per sample. Without the right models, trained on large, well-phenotyped cohorts and corrected for cell-type composition, age, and technical batch effects, those data points are noise.

HorizonBio's platform addresses this directly. We apply machine learning models trained on population-scale datasets to extract interpretable, clinically actionable signals from individual methylation profiles. Every prediction is accompanied by site-level explanations: the specific CpG loci driving the classification, so clinicians and patients understand not just the outcome but the mechanism behind it.

The longitudinal opportunity

The greatest opportunity in epigenomic medicine is longitudinal monitoring. A single test is a snapshot; repeated tests across time become a trajectory. That trajectory can confirm whether a lifestyle intervention is actually shifting biology, detect early signs of deterioration before symptoms appear, and personalise the timing and intensity of clinical follow-up.

This is the vision HorizonBio is working towards: not a report you file away, but a continuous conversation between molecular biology and clinical care, one that finally makes personalised medicine more than a promise.