← All research

Measuring Aging

Mapping how the body changes across a lifetime

Face at infancy Face in old age
Boy face at infancy Boy face in old age
One genome, a lifetime of change

The measurement problem

Aging is easy to recognise and hard to measure. Two people of the same age can differ by decades in how their tissues, organs and cognition are holding up, and no single number captures that difference. Our aim is to make aging quantifiable: to build measures of biological age that are cheap enough to repeat, precise enough to track one person over time, and mechanistic enough to point at what is actually going wrong.

Learning to measure

The data we need already exist — microscopy images, pathology slides, health records, biobanks — but they are far too large to score by hand. So we treat measurement as a machine-learning problem: assemble labelled data, train a model to predict age or a disease state from it, and test how well the prediction holds in cohorts the model has never seen. The workflow below is the one every biomarker in the lab passes through, from raw data to prediction.

Data collection
Data preparation
Model selection
Model training
Model evaluation
Model improvement
Prediction

The machine-learning workflow behind our senescence biomarkers — from data collection to prediction.

Senescence, read from the nucleus

The first target for that workflow is cellular senescence. As cells age they can enter a durable state in which they stop dividing yet remain metabolically active, secreting inflammatory signals that disturb the surrounding tissue. Senescent cells accumulate with age and contribute to pathologies from fibrosis to neurodegeneration, but identifying them has always been the bottleneck: the markers are inconsistent and the staining is laborious. Our models instead read senescence directly from the shape of the cell nucleus in ordinary microscopy images, with no specialised stain — and the same approach, applied to breast tissue biopsies, predicts future cancer risk.

Deep-learning pipeline detecting senescent cells from DAPI-stained nuclei
From a microscopy image of DAPI-stained nuclei, a deep neural network detects and segments individual nuclei, normalises each one for background, size and shape, and a second network reads the nuclear morphology to predict whether the cell is senescent.

From single cells to the whole phenome

Scaling up from the nucleus leads to the human aging phenome: the full spectrum of measurable traits — molecular, physiological, cognitive and clinical — that shift as people grow older. We mine health records, biobanks, imaging and multi-omics, together with tens of millions of histological samples in the human pathome, to map how thousands of age-related traits relate to each other, to sex, to tissue and to mortality. Aging clocks trained on these data flag when a person's biological age diverges from their chronological age, and the traits that diverge earliest become candidates for intervention.

The human pathome: from 33 million histological samples to age-modulating drugs
The human pathome (pathoage.com): 33 million histological samples and clinical pathology text are mined with machine learning to map thousands of age- and mortality-associated phenotypes, which are then cross-referenced in PubMed and tested as age-modulating interventions in cells and fruit flies.

Selected reading

  1. Ben Ezra M, Garbrecht JB, et al. & Scheibye-Knudsen M. The human pathome shows sex and tissue specific aging patterns. npj Aging (2026). doi:10.1038/s41514-025-00307-z
  2. Heckenbach I, Powell M, Fuller S, et al. & Scheibye-Knudsen M. Deep learning assessment of senescence-associated nuclear morphologies in mammary tissue predicts future risk of breast cancer. The Lancet Digital Health (2024). doi:10.1016/S2589-7500(24)00150-X
  3. Heckenbach I, Mkrtchyan GV, Ben Ezra M, et al. & Scheibye-Knudsen M. Nuclear morphology is a deep learning biomarker of cellular senescence. Nature Aging 2, 742–755 (2022). doi:10.1038/s43587-022-00263-3
  4. Teklu AA, Heckenbach I, Petr MA, et al. & Scheibye-Knudsen M. Deep learning reveals diverging effects of altitude on aging. GeroScience 47, 3873–3889 (2025). doi:10.1007/s11357-024-01502-8

See the full list of the lab's publications on Google Scholar.