Why accurately predicting age is not the same as measuring biological ageing
For more than a decade, DNA methylation clocks have transformed ageing research. By analysing patterns of DNA methylation across the genome, these algorithms can estimate a person's chronological or biological age with remarkable accuracy.
But there is an increasingly important scientific question:
The Core Question: What if predicting age isn't the same as measuring the biology of ageing?
Emerging research from the University of California, Berkeley suggests that while many current DNA methylation clocks excel at estimating age, they may overlook one of the most clinically relevant aspects of ageing: early biological instability.
The Challenge with Conventional Biological Age Clocks
Most first- and second-generation epigenetic clocks—including Horvath, Hannum, GrimAge, PhenoAge and DunedinPACE—are based on a machine learning approach known as Elastic Net regression.
Their goal is straightforward: find combinations of CpG methylation sites that best predict chronological age.
The result is an impressive mathematical model. The question is whether this is also the best biological model.
As Generation Lab explains, biological ageing is fundamentally non-linear. Human physiology does not age at a constant rate. Instead, people experience long periods of stability interrupted by episodes of accelerated molecular change driven by inflammation, stress, disease, injury and recovery.
Biology Doesn't Move in One Direction
When ageing or disease develops, thousands of DNA methylation sites (CpGs) begin changing simultaneously. However, they do not all change in the same direction.
- Some CpG sites become hypermethylated.
- Others become hypomethylated.
- Both are biologically meaningful.
Each reflects different regulatory pathways responding to cellular stress, inflammation and loss of homeostasis.
This is where an important limitation of conventional clocks begins to emerge.
When Biology Gets Averaged Away
During model training, Elastic Net assigns each CpG a mathematical weight. Some weights are positive; others are negative. The final biological age estimate is calculated by summing all weighted CpG signals.
The Problem: Positive signals + negative signals = one final age prediction. Opposing biological responses can effectively cancel one another out.
Instead of reinforcing evidence of biological dysfunction, the model averages many of these changes away.
Researchers at UC Berkeley describe this phenomenon as a decoherence effect, where opposing disease-related methylation changes reduce the model's ability to distinguish biological dysregulation from healthy ageing.
Why This Matters Clinically
Imagine two 55-year-old individuals. One is healthy. The other has chronic low-grade inflammation and early molecular dysfunction that has not yet produced symptoms.
Traditional DNA methylation clocks may estimate both individuals to have very similar biological ages—not because their biology is the same, but because the underlying molecular changes have been averaged into a similar overall prediction.
The result is a clock that predicts age accurately while potentially missing important early biological changes associated with disease initiation.
The Evidence
In their recent GeroScience publication, Skinner, Conboy and Conboy evaluated several widely used first- and next-generation DNA methylation clocks. Among their findings:
- CpGs with the greatest biological relevance were often not the most influential features selected by Elastic Net models.
- Conventional clocks showed limited ability to distinguish healthy individuals from patients with chronic inflammatory diseases.
- Prediction residuals for healthy individuals commonly varied by ±10–20 years, making it difficult to interpret modest differences in biological age as clinically meaningful.
- Disease-related methylation changes frequently cancelled one another within linear models, reducing their ability to detect inflammaging.
In other words: the clocks often estimated chronological age well—but were less effective at detecting the biological processes driving ageing.
A Different Approach: Feature Rectification
Rather than allowing opposing methylation signals to negate each other, Professor Irina Conboy's group developed an approach called feature rectification.
The concept is simple but powerful. Before modelling, biologically meaningful methylation changes are aligned so that disease-related signals reinforce one another instead of cancelling out.
The Shift in Goal: From predicting chronological age — to — quantifying the cumulative burden of biological dysregulation.
Using this approach, the researchers demonstrated improved separation between healthy individuals and patients with inflammatory diseases, suggesting that coherent biological signals may provide a more informative measure of ageing biology than traditional age prediction alone.
From Biological Age to Biological Instability
This work reflects a broader shift occurring within longevity medicine. Historically, the question has been: "How old are you biologically?"
Increasingly, clinicians are asking:
- Which biological systems are losing resilience?
- Where is molecular regulation beginning to fail?
- Can we detect dysfunction before conventional biomarkers become abnormal?
- Can interventions restore biological stability before disease develops?
These questions move beyond age estimation toward measuring the biological processes that drive ageing itself.
How SystemAge Builds on This Science
SystemAge™ was developed around this emerging scientific framework. Rather than relying solely on conventional biological age clocks, SystemAge™ quantifies epigenetic noise—a measure of molecular instability across biologically important CpG sites—and evaluates patterns of dysregulation across multiple functional systems.
According to Generation Lab, increased epigenetic noise reflects a loss of biological homeostasis that often precedes measurable functional decline, and the platform resolves this information across 21 functional regulatory systems to provide more granular biological insight.
The objective is not simply to estimate how old someone appears biologically. It is to identify where biological resilience may already be declining, potentially before structural pathology or clinical symptoms emerge.
The Future of Longevity Diagnostics
The next generation of longevity diagnostics is unlikely to be defined solely by increasingly accurate age prediction. Instead, it is moving toward quantifying:
- Biological instability
- Loss of molecular resilience
- Organ- and system-specific dysregulation
- Longitudinal response to interventions
This represents a transition from predicting time to measuring biology. For clinicians, researchers and patients alike, that distinction may prove far more valuable.
Key Takeaway: The future of precision longevity medicine is not simply about estimating biological age—it is about detecting the earliest signs of biological instability, enabling intervention before dysfunction becomes disease.
Sources:
DNA methylation clocks struggle to distinguish inflammaging from healthy aging, but feature rectification improves coherence and enhances detection of inflammaging. Skinner CM, Conboy MJ, Conboy IM. Geroscience. 2025 Jun;47(3):3043–3060. doi: 10.1007/s11357-024-01460-1. Epub 2025 Jan 18. PMID: 39825170; PMCID: PMC12181618.
