Live·Open questions in longevity research
All news
Longevity research

The method that identified two peaks in molecular aging at ages 44 and 60 in a Nature Aging paper produces similar peaks from random numbers: the paper’s authors tested this with Harvard biostatisticians in a preprint

12 November 2026· 261112004

The method that identified two peaks in molecular aging at ages 44 and 60 in a Nature Aging paper produces similar peaks from random numbers: the paper’s authors tested this with Harvard biostatisticians in a preprint

On November 10, Harvard biostatisticians released the second version of a preprint, a manuscript that has yet to undergo peer review. Its authors include Michael Snyder and Xiaotao Shen, the research lead and first author of the 2024 paper.

In August 2024, Stanford University School of Medicine reported two periods of rapid change in the body, around ages 44 and 60. In the Nature Aging paper, Snyder’s group measured about 135 thousand molecules and microbes in 108 people aged 25 to 75. The number of molecules that changed with age peaked twice; these peaks were described as “waves of aging.”

The peaks were identified using DE-SWAN. This method moves a 20-year window along the age range and counts molecules whose levels differ significantly between the younger and older halves of the window, meaning that the differences are unlikely to have arisen by chance. Tony Wyss-Coray’s group at Stanford introduced DE-SWAN in 2019. The study had few younger participants: only eight were aged 25 to 40. According to Shen’s email to GenomeWeb, the team was concerned that DE-SWAN would lack sensitivity, so they smoothed the data using LOESS. For each molecule, they fitted a smooth curve whose values were calculated from nearby measurements, then sampled the curve at half-year age intervals.

John Quackenbush’s laboratory at Harvard tested the procedure using data and code supplied by Snyder and Shen. The Harvard researchers generated random numbers unrelated to age for a thousand hypothetical molecules, then applied smoothing and DE-SWAN using four sets of ages. In every window, most “molecules” were classified as changing significantly. When the researchers used the ages of Snyder’s participants, the number of “significant” molecules rose twice, in the early forties and early sixties.

The test comparing the two halves of the window assumes independent measurements, but receives points from a single smooth curve. If the curve trends in one direction, nearly all the values in one half of the window are lower than those in the other, and the test detects a difference. Where participants are sparse across the age range, the curve becomes more distorted, and those windows contain more “significant” molecules. When the original paper’s gene activity data were analyzed without smoothing, no gene showed a significant difference. Even without smoothing, however, DE-SWAN produces “waves” in a simulation where every molecule changes at a constant rate: the number of “significant” molecules depends on the number of participants in the window.

The 2024 paper checked the peaks by shuffling ages: real peaks should disappear when ages are scrambled. The peaks disappeared, but the authors had shuffled curves that were already smoothed. Points on such a curve depend on one another and cannot validly be shuffled. When the Harvard researchers shuffled ages before smoothing, the peaks remained. Shen acknowledged: “From our original analysis alone, it is difficult to distinguish how much of the observed waves reflects genuine biological signals and how much may result from the statistical procedure itself.”

On July 15, 2026, the editors of Nature Aging added a note to the paper stating that the reliability of some analyses used to identify the peaks was “currently in question.” Alan Cohen’s group at Columbia University independently concluded that DE-SWAN is unreliable. The preprint’s authors estimate that 63 aging papers citing the 2019 study use LOESS or DE-SWAN; 25 of these were published in 2026 (as of September). The authors question the conclusions of these papers.

In August, Snyder told GenomeWeb: “The part about nonlinear aging definitely still holds.” The preprint’s authors leave open the possibility that nonlinear aging could be found in other samples using other methods.

Originally published on Telegram by Ukhvat NewsView on Telegram
Sources
#molecular-aging#de-swan#loess-smoothing#statistical-artifacts#aging-waves