Why Your Biological Age Score Might Be Wrong

The paper, by Mei and colleagues, is not a tabloid takedown. It is a careful methodological critique of the machine-learning logic that underpins the most widely used DNA methylation "age clocks" — the kind of model that powers many of the at-home tests now marketed to health-curious consumers. Their central claim is uncomfortable: a clock can be excellent at predicting chronological age and still be a poor instrument for understanding biological aging. The reason is that the math rewards features that move in a clean line from birth toward death, even when the underlying biology does nothing of the sort.

Aging, the authors remind us, is not linear. It has different rhythms in childhood, midlife, and the years after menopause, with abrupt inflection points rather than a tidy slope. When an algorithm is trained to compress that messy reality into a single rising line, it tends to keep the features that cooperate and quietly discard the ones that don't — even if the discarded signals are the biologically interesting ones. The output looks impressive. The interpretability suffers.

One of the paper's sharpest observations concerns what the authors call incoherence. In the DNA methylation patterns these clocks read, some sites gain methylation as we age and as disease accumulates; others lose it in the same direction. A biologically faithful model would treat those two groups as opposites. Yet the authors show that major conventional clocks assign positive weights to both kinds of sites, effectively adding signals that should be subtracting from one another. The model still produces a number. It just isn't the number you think it is.

The consequences are not abstract. The same analysis finds that popular clocks are skewed toward leukocyte fractions — heavily influenced by the mix of white blood cells in a sample rather than by aging itself. A score can shift because your immune cell composition shifted that morning, not because anything meaningful changed about your biological trajectory. For a reader weighing whether last quarter's number really reflects last quarter's habits, that is a sobering caveat.

Perhaps the most striking finding, especially for women navigating the post-menopausal decade, concerns inflammaging — the chronic, low-grade inflammation now considered a hallmark of biological aging and a driver of cardiovascular, metabolic, and cognitive decline. The authors argue that conventional age clocks struggle to detect inflammaging, precisely because the mathematical scaffolding that makes them look accurate also flattens the biological signal that inflammation leaves behind.

When the authors rebuild the models to remove the incoherence — forcing the math to respect the direction biology is actually moving — the picture changes. The corrected models are less skewed toward neutrophils and better at detecting inflammaging. That is not a marketing flourish; it is a methodological proof of concept that the standard approach is leaving important information on the table.

None of this means biological-age testing is meaningless, and the GeroScience authors do not argue that it is. They are clear-eyed about the appeal of a single, legible score, and they sketch a path forward through non-linear machine-learning approaches that may better honor the shape of aging. The point is more measured: today's commercial tests rest on a methodological tradition whose limitations are now being articulated in the peer-reviewed literature, and the informed consumer should know that.

For readers who have already taken a test, the practical reframing is this. A biological-age score is best understood as one signal among many — interesting, not authoritative, and likely sensitive to factors (like the immune cell mix in a given blood draw) that have little to do with how you have been living. A score that drops after a quarter of better sleep and walking is encouraging; a score that jumps after a stressful month is not a diagnosis. Anyone weighing what a result means for their actual health decisions should talk it through with a clinician who can place it alongside the rest of the picture — blood pressure, lipids, inflammatory markers, function, history.

The longevity field is moving quickly, and the candor of this paper is itself a good sign. The same researchers building the next generation of clocks are publicly auditing the last one. That is how a young science matures — not by retreating from the promise of measuring aging, but by being honest about which measurements have earned our trust, and which are still a work in progress.