Proteomic aging clocks: how blood-protein age estimates are built

Prepared by: Research Desk. Sources are linked for verification.

Most biological-age numbers in the quantified-self world come from DNA methylation. A second family of aging clocks reads a different molecular layer: the thousands of proteins circulating in blood plasma. A proteomic aging clock is a statistical model that estimates age from the levels of a panel of plasma proteins. Since 2019, several large human studies have built these models and linked their output to disease and mortality records. This explainer covers how the clocks are built, what the key studies reported, how they differ from epigenetic clocks, and where their limits lie.

What a proteomic aging clock is

The basic logic matches any aging clock. Researchers measure many molecules in a large group of people of known age. They then fit a model that predicts age from those measurements.

For a proteomic clock, the inputs are protein abundances measured in plasma. The output is a "protein age." The quantity most studies care about is the protein age gap: predicted protein age minus chronological age. A positive gap means the protein profile looks older than the calendar age, according to that model. A negative gap means it looks younger.

The model is only as general as the data it was trained on. A protein age gap is therefore a statement about how one person's plasma compares with a specific training population, measured on a specific platform.

How the proteins are measured

Two high-throughput technologies dominate this research. They work differently.

• Aptamer-based assays (SomaScan). These use modified short DNA-like molecules (aptamers) that bind target proteins. The bound aptamers are then counted. The Lehallier 2019 study used this approach [1].

• Proximity extension assays (Olink). Pairs of antibodies carrying DNA tags bind the same protein. When both bind, the tags come close enough to be joined and amplified. The UK Biobank clock by Argentieri and colleagues used this platform [3].

The two technologies do not target identical protein sets. Even when they target the same protein, they can capture it differently. A clock trained on one platform is not interchangeable with one trained on the other. This is one of the biggest practical constraints in the field.

How a proteomic clock is built

The workflow resembles the one described in how epigenetic clocks work:

1. Assemble a cohort. Collect plasma from a large number of people across a wide age range, with linked health records.

2. Measure the proteome. Quantify thousands of proteins per sample on one platform.

3. Select predictors. Use penalized regression or machine learning to find a smaller subset of proteins that jointly predict age.

4. Validate. Test the model on held-out samples, and ideally on separate cohorts from different populations.

5. Relate the gap to outcomes. Check whether the protein age gap is associated with later disease or death, after accounting for chronological age.

Step 5 separates a clock that only tracks the calendar from one that captures risk-relevant variation. The same distinction shaped the move from first- to later-generation methylation clocks, covered in epigenetic clock generations.

What the key studies reported

Lehallier 2019: waves of change across the lifespan

Lehallier and colleagues measured 2,925 plasma proteins in 4,263 people aged 18 to 95 [1]. They reported that protein changes with age were not smooth. Instead they clustered in waves, with notable shifts around ages 34, 60 and 78 [1]. A set of 373 proteins predicted chronological age [1]. NIH's summary of the work described these blood protein signatures as changing across the lifespan [2]. This undulating pattern matters for interpretation: a protein that moves sharply in one decade may barely move in another.

Argentieri 2024: a UK Biobank clock tested in other populations

Argentieri and colleagues measured 2,897 proteins in 45,441 UK Biobank participants using the Olink platform [3]. They built a 204-protein clock. Its predicted age correlated with chronological age at r = 0.94 [3]. The model was then tested in two further cohorts:

A larger protein age gap was associated with 18 major chronic diseases, with multimorbidity and with all-cause mortality [3]. The study is observational. It reports associations within cohorts, not a mechanism or a tool validated for individual decisions.

Oh 2023 and later work: organ-specific protein age

Oh and colleagues took a different angle. They used proteins whose expression is enriched in particular organs to estimate separate "organ ages." In 5,676 people, they derived aging signatures for 11 organs and reported that these were associated with organ-related disease risk and mortality [4].

Later studies extended this idea. Goeminne and colleagues built plasma-protein models of organ-specific aging and mortality. They framed diseases as accelerated aging of particular organismal systems [5]. In the Whitehall II cohort, Kivimäki and colleagues examined proteomic organ-specific aging signatures against 20-year risk of age-related diseases [6]. A 2025 study by Oh and colleagues linked brain and immune system aging, estimated from plasma proteomics, with healthspan and longevity [7].

How proteomic clocks differ from DNA methylation clocks

Both are "aging clocks," but they read different biology.

• Molecular layer. Methylation clocks read chemical tags on DNA, as in the multi-tissue Horvath clock [8]. Proteomic clocks read the abundance of circulating proteins, which sit closer to the functional state of tissues.

• Training target. Some methylation clocks were trained on chronological age [8]. Others were trained on mortality-linked composites, as with GrimAge [9], or on the pace of physiological decline, as with DunedinPACE [10]. Most proteomic clocks described above were trained on chronological age and then tested against outcomes. See GrimAge explained for how a mortality-trained design differs.

• Organ resolution. The organ-enriched protein approach offers a way to estimate several system-level ages from one blood draw [4][5]. Standard methylation clocks from blood give a single summary number.

• Sample type. Proteomic clocks require plasma. Many consumer methylation tests use saliva or dried blood spots.

Neither approach is established as a replacement for the other. Research is also moving toward combining layers. OMICmAge, for example, integrates multi-omics data with electronic medical records [13]. For how methylation clocks fared against mortality in cohort studies, see epigenetic clocks and mortality prediction.

Limitations to keep in mind

• Platform dependence. Results from aptamer-based and antibody-based assays are not directly interchangeable [1][3]. A protein age from one model cannot be read on another model's scale.

• Population generalizability. Validation in Chinese and Finnish cohorts is encouraging [3], but most training data still come from a limited set of large biobanks.

• Association, not causation. A larger protein age gap is linked with higher disease and mortality risk at the group level [3][4]. That does not show that shifting the gap would change risk.

• Non-linear change. The wave-like pattern reported by Lehallier means a given protein's movement depends heavily on life stage [1].

• Regulatory status. A research clock is not the same as a validated diagnostic. Tests developed and run within a single laboratory are generally offered as laboratory-developed tests [11]. FDA's 2024 rule covering these tests was vacated, and FDA published its implementation of that vacatur in September 2025 [14]. CLIA certification addresses laboratory quality standards [12]. It is not proof that an aging score is clinically meaningful. Our page on whether biological age tests are FDA approved covers this in detail.

Further reading on lab testing

Readers weighing any biological-age test can apply the same neutral criteria. Which molecular layer is measured (methylation, plasma proteins or standard blood chemistry)? Which published model is the report based on? Is the measurement platform named? Is the laboratory CLIA-certified? Are the full price and the cost of retests stated up front? What happens to the sample and the data afterward? The biological age test guide explains how to interpret a result once it arrives.

Any program listed here is judged on the criteria above: what it measures, how it reports validation, price transparency and data privacy.

Frequently asked questions

What is a protein age gap?

It is the difference between the age a proteomic clock predicts from plasma proteins and a person's chronological age. In the UK Biobank study, a larger gap was associated with 18 major chronic diseases and all-cause mortality at the group level [3].

Are proteomic clocks more accurate than epigenetic clocks?

The two measure different molecular layers and were often trained toward different targets, so a single accuracy ranking is not meaningful. The UK Biobank proteomic clock correlated with chronological age at r = 0.94 [3]. Correlation with age is only one property, though. Association with outcomes and repeatability also matter.

Can results from different protein platforms be lined up?

Not directly. Aptamer-based and antibody-based assays measure proteins differently, so each clock applies only to data from the platform it was built on [1][3].

Are proteomic aging clocks used as diagnostic tests?

The studies described here are research models from observational cohorts. Being run in a CLIA-certified lab does not by itself establish clinical validity [12].

Related on this site

• How old are you, biologically?

• How epigenetic clocks work: DNA methylation and age estimation

• First, second and third generation epigenetic clocks, explained

• Epigenetic clocks and mortality prediction: what cohort studies found

• GrimAge explained: how the mortality-trained epigenetic clock works

• About BioAge Peptide: how this site works and disclosures

Last updated: 25 October 2026.

Not medical advice. This page reports published research and regulatory records for information only.

Research-only notice: Content on this site is provided for research and educational purposes only and is not intended for human use. It is not medical advice.

Translation notice: Some content on this site may have been machine-translated using Google Translate and may contain errors.