Question 1

Should a new clock be built, or existing ones applied?

Accepted Answer

Building a new predictor is justified when the tissue, species or training outcome is not covered by existing models, or when the training target itself is the contribution. Otherwise, applying established clocks and interrogating what drives their residuals often makes a stronger thesis, since it puts the biological question rather than the regression at the centre.

Question 2

Why does cell type composition matter so much?

Accepted Answer

Because bulk methylation from blood or tissue reflects the mixture of cells present as much as any change within them. Composition shifts across the lifespan can produce an apparent epigenetic signal that is really a counting effect. Deconvolution estimates or measured cell counts belong in the model, with results reported both with and without adjustment.

Question 3

What separates early clocks from later generations?

Accepted Answer

Early models were trained to predict chronological age, so accuracy was the objective and health-relevant variance was treated as error. Later models train on mortality risk, physiological composites or longitudinal rate of change, which usually weakens age correlation while strengthening association with outcomes. The choice should follow the research question.

Question 4

How is age acceleration properly defined and analysed?

Accepted Answer

Usually as the residual from regressing predicted on chronological age within the analysis sample, which removes the built-in correlation by construction. Analyses then treat that residual as exposure or outcome with appropriate covariates. Using the raw difference instead reintroduces age dependence and is a frequent source of confusion in reported results.

Question 5

What data governance applies to cohort methylation data?

Accepted Answer

Access generally requires a data access agreement, ethical approval covering the intended analysis, and adherence to storage and sharing conditions set by the cohort. Genotype-adjacent information can be inferred from arrays, so re-identification risk is real. Application timelines are often long and should begin before the analysis plan is finalised.

Question 6

Which normalisation approach should be used?

Accepted Answer

No single approach is correct, but the choice should be made once, justified, and applied consistently. Probe type bias, background correction and batch adjustment all shift downstream site selection. Sensitivity of the final model to normalisation is worth reporting, since examiners often ask whether conclusions survive an alternative processing pipeline.

Question 7

Can a clock trained in one tissue be applied to another?

Accepted Answer

Generally not without evidence. Site-level signatures differ substantially across tissues, and transfer usually degrades accuracy in ways that are far from uniform. Either use a predictor trained in the relevant tissue, or demonstrate performance there first, treating cross-tissue application as a hypothesis to be tested rather than an assumption.

Biotechnology

Biotechnology

Research Frontiers in Epigenetic Aging Clock Development