SARDINE Converts EWAS Data Into Biomarker-Level Signals Without Individual Records
Epigenetic clocks have become one of the most widely discussed tools in aging research, yet their clinical meaning has stayed stubbornly unsettled. The new SARDINE framework aims to change the way those biomarkers are evaluated by extracting clock-phenotype associations from EWAS summary statistics, clock weights, and reference methylation covariance, all without needing individual-level data.
That shift matters because EWAS datasets are accumulating far faster than the field can reinterpret them at the biomarker level. CpG-by-CpG results are useful for discovery, but they are difficult to translate into the behavior of a composite epigenetic clock, especially when those clocks are meant to summarize biological aging rather than isolated methylation changes at single loci. SARDINE is designed to bridge that gap by converting summary statistics into interpretable biomarker-level associations.
Performance Across 38 Epigenetic Biomarkers and Four Phenotypes
Investigators evaluated the framework across 38 epigenetic biomarkers and four phenotypes in 15,677 individuals. Across that benchmark, SARDINE accurately reproduced individual-level association estimates. The result is important because it shows that biomarker-level inference can be reconstructed from aggregated data rather than reserved for cohorts with full participant-level access.
A single computational layer can now reuse data that would otherwise sit idle after the original EWAS publication.
The practical implication is straightforward. Large collections of epigenetic association results can be revisited in a standardized way, allowing researchers to ask whether a clock behaves consistently across aging-related traits, disease states, and lifestyle exposures. In a field where one biomarker may appear promising in one dataset and unconvincing in another, a method that can recover comparable effect estimates from summary data could reduce the noise created by incompatible analytic pipelines.
Heterogeneity Across Aging, Disease, and Lifestyle Associations
When SARDINE was applied to diverse EWAS datasets, biomarker associations showed substantial heterogeneity across aging, diseases, and lifestyle factors. That pattern suggests the clocks do not move as a uniform biological chorus. Some may track broad aging processes, while others may respond more sharply to specific exposures or pathophysiologic states.
This is where the biology becomes especially interesting. Epigenetic clocks are often treated as shorthand for systemic aging, yet methylation patterns can reflect very different upstream pressures: inflammation, metabolic stress, immune remodeling, smoking-related injury, medication effects, or disease-related tissue changes. A biomarker that correlates with one phenotype may not do the same for another, because clock weights can amplify some CpG signals while muting others. SARDINE makes those differences more visible by translating CpG-level EWAS output into a higher-order biomarker readout.
The framework therefore does more than economize analysis. It changes the unit of interpretation. Instead of treating every EWAS as an endpoint in itself, the method allows investigators to map the same summary data onto a biomarker that may better approximate the biological construct of interest.
Why Summary Statistics Matter for Aging Research
Much of the field’s epigenetic evidence remains trapped inside summary outputs. That creates a familiar problem in biomedical research: the data exist, but the analytic window is too narrow to compare studies cleanly or to reuse them for new questions. SARDINE addresses this bottleneck by relying on three ingredients already common in published EWAS work: summary statistics, clock weights, and a reference methylation covariance structure.
The computational appeal is substantial. Individual-level methylation data can be difficult to share because of privacy, governance, and institutional restrictions. Summary statistics are much easier to distribute, but they have traditionally limited the depth of downstream biomarker analysis. By showing that clock-phenotype association estimates can be recovered from those summaries, SARDINE offers a scalable route for meta-analytic aging research and cross-study biomarker comparison.
For translational scientists, the message is not that every epigenetic clock is ready for bedside use. It is that biomarker evaluation can become more systematic, more reproducible, and less dependent on privileged access to raw participant data. In a field crowded with proposed aging measures, the ability to re-score existing EWAS literature may separate durable signals from one-off findings.
Clinical Interpretation Remains Indirect, But the Analytical Gain Is Real
No direct patient-level intervention follows from this preprint, and the framework does not claim to prove causality. Its contribution sits one layer upstream, in the architecture of biomarker inference. Even so, the method could influence how future aging studies are designed and interpreted, because it allows researchers to compare clocks across traits using a consistent statistical bridge.
For clinicians and investigators watching the aging biomarker space, the most relevant consequence is the possibility of cleaner prioritization. If a clock repeatedly shows association across multiple EWAS datasets, that pattern may justify deeper mechanistic work. If it fails to reproduce once transformed from CpG-level data to biomarker-level estimates, the apparent signal may have been more fragile than it first looked.
SARDINE also offers a useful reminder about epigenetic biomarkers in general. A clock is not a single pathway; it is a weighted summary of many methylation sites, and its behavior depends on how those sites covary in the population being studied. Methods that ignore that structure risk flattening biology into a crude average. Methods that use it can recover a more faithful representation of the underlying biomarker.
The Next Test Is External Reproducibility Across Broader Cohorts
The preprint’s strongest claim is technical and empirical: SARDINE reproduced individual-level association estimates in a sizable benchmark and then revealed heterogeneous biomarker associations across EWAS datasets. What remains unresolved is how broadly that performance will hold across additional cohorts, different methylation platforms, and future biomarker definitions.
Those questions are especially important because the epigenetic clock field changes quickly. New weights, new reference panels, and new phenotype targets can all shift how a composite biomarker behaves. For now, the method provides a scalable analytic bridge, but its value will ultimately depend on whether other groups can use it to reproduce the same associations in independent datasets and whether those reconstructed signals hold up when the biology gets more complex than the current benchmark allows.