How Every Cell Ages: A Journey Through the Methylation Landscape express
DNA methylation consists of tiny chemical marks on our genes. Over the years, they change on a fairly strict schedule. Standard epigenetic clocks look at these marks and output a single number: biological age. But a number can't tell you how the pattern of marks rearranged day by day, year by year. The authors of this work decided to turn a static photograph into a continuous film.
To do this, they collected methylation data from thousands of people of different ages — from infants to centenarians. Then they built a mathematical landscape. Each point in it corresponds to a specific pattern of marks, and age sets the direction of movement. First, a special autoencoder (an algorithm that compresses complex data into a simple map) turned the varied pattern into a compact scheme while preserving the ability to decode it back. Then, an optimal transport model calculated the most natural aging path through this landscape. It accounted for both random walks and the inevitable loss of cells with age.
The most striking result appeared after age 60–70. The rate of random changes jumps sharply. The aging landscape seems to become more 'loose,' and the trajectories of different people diverge more and more. Reverse decoding showed that these divergences are not chaotic. They coalesce into several typical scenarios — archetypes of aging, so to speak. Now they can be studied individually.
Thus, from a static number on a 'clock,' science moves to mapping molecular life. And for the first time, this map has a relief — with accelerations, decelerations, and forks. These are yet to be explored.
In newborns, the DNA methylation pattern is almost identical, but by age 80, differences between people become so large that these marks can reliably distinguish one person from another, like fingerprints, but molecular.
In the series 'Altered Carbon,' wealthy people store backup copies of their consciousness on a 'stack' but can't stop the biological shell from aging. The model from this work is a step toward learning not just to measure the shell's wear and tear but to predict and correct its trajectory.