A single blood draw may reveal which cell types are aging fastest, offering a new window into disease risk long before symptoms emerge.

Spotlight: Translating cellular aging clocks into disease risk prediction. Image Credit: AtlasStudio / Shutterstock

In a recent spotlight review published in the journal Cell Reports Medicine, researchers evaluated a landmark study by Ding et al. that developed machine-learning-based cellular aging clocks readable from human blood.

The novel framework was used to map more than 7,000 circulating plasma proteins to more than 40 cell types using Human Protein Atlas data, while the aging clocks were developed and validated across nearly 60,000 individuals. The framework demonstrated that, at the molecular level, biological aging varies markedly across individual cell types.

More importantly, the study demonstrated that these cell-specific aging trajectories could be used to predict the risk of neurodegenerative disorders, malignancies, and all-cause mortality. Notably, some disease associations were observed over a follow-up period extending to 15 years, highlighting the framework’s potential for future clinical risk stratification rather than for immediate clinical use.

The review’s authors highlight that tracking dynamic, cell-resolved proteomic signatures provides a functional layer of risk stratification that can add information beyond conventional static genetic assessments in some settings.

Background

Geroscience has long established that chronological age fails to capture the true biological heterogeneity of human aging. Unfortunately, reviews of field data indicate that, despite individuals of the same age often exhibiting disparate clinical trajectories, conventional tools have historically relied on static genetic instruments (e.g., polygenic risk scores [PRS] and familial carrier statuses) to estimate inherited disease risk rather than dynamic biological aging.

However, recent research suggests that although genetic markers encode baseline vulnerability, they cannot capture time-varying physiological decline. These conventional approaches are further limited by their inability to recognize or quantify whether specific cell populations are aging prematurely or at rates significantly different from the body’s average.

Advances in plasma proteomics have begun to address this limitation by providing dynamic molecular measures of aging. The reviewed study’s key advance was to treat circulating proteins as signals from specific cell types rather than solely as generic systemic biomarkers. Consequently, the medical community has, to date, lacked a non-invasive methodology capable of resolving plasma proteomic aging signals at this level of cell-type specificity.

About the study

The present study aimed to address these empirical limitations and inform future precision medicine by developing a novel machine-learning (ML)-based framework capable of constructing high-resolution cellular aging clocks from a single blood draw.

The ML models were trained to compute biological age (cell-type-specific) based on the abundance of cell-enriched proteins. The approach was validated across the Global Neurodegeneration Proteomics Consortium (GNPC; n = 14,281), the UK Biobank (UKB; n = 44,458), and the 1946 National Survey of Health and Development (NSHD; n = 1,803), with replication across both SomaScan and Olink proteomic platforms.

Using transcriptomic reference data from the Human Protein Atlas (HPA), the researchers mapped more than 7,000 circulating plasma proteins to cell types showing enriched expression of their corresponding genes. The cell types themselves were found to span more than 40 distinct types.

The researchers then examined whether these cellular aging signatures were associated with long-term incidence of neurodegenerative disease, lung cancer risk, and graded mortality risk using a composite Polycellular Aging Risk Score (PARS).

Study findings

The study’s analyses established that the cell-specific proteomic aging profiles identified by the study’s ML-based platform were strongly associated with future disease risk.

The most notable of these associations is that of extreme astrocyte aging, which was identified as the single strongest predictor of incident Alzheimer’s disease (AD) in the UKB cohort (hazard ratio [HR] of 12.59 over 15 years of follow-up). In comparison, established AD risk measures including APOE4 carrier status, AD polygenic risk scores, and chronological age were associated with HRs of 5.30, 2.14, and 1.24, respectively.

Furthermore, the study found that among high-risk APOE4/4 homozygotes, cumulative AD incidence was established at 38.3% in individuals with extreme astrocyte aging, compared to only 12.6% in normal agers and 0% among those with youthful astrocyte profiles.

When stratifying the sample cohort by sex, the analyses revealed heightened vulnerability in women possessing both APOE4 and extreme astrocyte aging (HR = 14.23, 95% CI: 9.86–20.54) compared to men (HR = 10.95).

The platform similarly demonstrated that inhibitory (not excitatory) neuron aging was specifically associated with AD, that extreme skeletal myocyte aging was associated with a 12.7-fold increased risk of amyotrophic lateral sclerosis (ALS; HR = 12.7), an association that persisted when analyses were restricted to cases diagnosed more than three years after blood collection.

Extreme alveolar and respiratory epithelial cell aging also amplified lung cancer risk by 58% above smoking alone, while alveolar cell aging was associated with lung cancer with an HR of 8.39. Myeloid-lineage aging also identified normoglycemic individuals at elevated risk of type 2 diabetes.

Finally, the study’s PARS polycellular evaluations revealed that 15-year survival declined from ~90% in normal agers to ~34% among individuals exhibiting extreme aging across 20+ cell types.

Conclusions

The reviewed study advances precision medicine by showing that cell-type-specific aging clocks derived from plasma proteins could provide clinically informative assessments of disease risk across multiple systems.

While the study was notably limited by the predominant representation of older European cohorts, probabilistic cell-type mapping based on two-fold transcriptomic enrichment thresholds, possible reverse causation in cross-sectional analyses, and the need for orthogonal validation using cerebrospinal fluid biomarkers, positron emission tomography, or neuropathology, it represents an important step towards determining whether these proteomic aging clocks can eventually support clinical practice.

Prospective validation, replication in more diverse populations, mechanistic investigation, and regulatory evaluation remain necessary before clinical-grade diagnostic deployment.

Journal references:

• Heikal, S., & Salama, M. (2026). Translating cellular aging clocks into disease risk prediction. Cell Reports Medicine, 7(8), 102996.DOI:10.1016/j.xcrm.2026.102996. https://www.cell.com/cell-reports-medicine/fulltext/S2666-3791(26)00413-1
• Ding, D. Y., Bot, V. A., Chen, K. L., Groves, J. W., Pálovics, R., Masuda, D., Farinas, A., Oh, H. S. H., Wagner, V., Lu, N., Cruchaga, C., Isakova, A., Schott, J. M., & Wyss-Coray, T. (2026). Plasma proteomic signatures of cellular aging predict human disease. Nature Medicine,32(6), 2060-2072.DOI: 10.1038/s41591-026-04446-y, https://www.nature.com/articles/s41591-026-04446-y