Sample Submission Guidelines
Which Methylation Platform Fits Your Clock Objective
Every clock project begins with the same practical question: what methylation data do I need, and how much coverage is sufficient for reliable age prediction? The answer depends on whether you are discovering new age-associated CpGs, applying an existing clock to your cohort, or validating a defined marker set. The table below maps each platform to the clock use case it serves best, along with the trade-offs that affect study design and budget planning.
| Platform | Coverage Profile | Best Fit for Clock Research | Practical Considerations |
|---|---|---|---|
| Whole-Genome Bisulfite Sequencing (WGBS) | Genome-wide, single-base resolution covering all CpG, CHG, and CHH contexts | First-generation clock discovery, multi-tissue models, cross-species clock development, and projects where novel age-associated loci must be identified rather than assumed | Higher sequencing cost and DNA input (500 ng – 1 µg); provides the most comprehensive methylome baseline for models that do not presuppose which CpGs matter |
| Reduced Representation Bisulfite Sequencing (RRBS) / EM-seq | CpG-dense regulatory regions — promoters, CpG islands, enhancers — enriched by restriction digestion (RRBS) or enzymatic conversion (EM-seq) | Regulatory-region-focused clock profiling, promoter-centric aging models, and projects needing a cost-efficient balance between breadth and per-sample sequencing depth | Lower coverage outside CpG islands; EM-seq offers gentler conversion chemistry than bisulfite, improving data from lower-quality DNA starting material |
| Targeted Bisulfite Sequencing Panels | Selected known clock CpGs or custom age-associated marker sets at ultra-high sequencing depth | Validation studies, large cohort screening, cfDNA-based clock research, and cost-effective deployment of custom marker panels when the clock CpGs are already established | Requires pre-existing marker list; highest cost efficiency for cohort-scale screening; ultra-deep coverage enables confident detection at sites with partial methylation |
| DNA Methylation Arrays (Infinium EPIC v2 / 935K) | 935,000+ predefined CpG probes with standardized, well-annotated content optimized for human methylation analysis | Compatibility with established array-trained clocks (Horvath, Hannum, GrimAge, DunedinPACE), large clinical or epidemiological cohorts, cross-study meta-analyses, and longitudinal comparisons | Fixed probe content limits discovery of novel markers; highest standardization across batches and laboratories; extensive published clock models built on the Infinium platform |
| Native Long-Read Methylation Profiling (Nanopore) | Single-molecule methylation detection across long DNA fragments without bisulfite conversion, preserving native base modifications | Haplotype-resolved methylation analysis, repetitive-region clock discovery, combined genetic–epigenetic interrogation from a single assay, and projects where bisulfite-induced degradation compromises data quality | Higher per-base error rate relative to short-read bisulfite methods; best applied when long-range methylation context (haplotype phasing, transposon silencing, satellite repeat methylation) addresses a specific biological question |
Clock Research Scenarios We Support
Epigenetic clock research spans a wide methodological landscape — from applying established human clocks in clinical cohorts to developing novel multi-species aging models. The right platform, QC strategy, and analysis pipeline vary with each scenario. Below we describe the configurations we support and the decisions we make at project start to match the workflow to the research question.
Biological age comparison across treatment and disease groups
- Profile methylation levels across case–control, longitudinal, or intervention cohorts to compare biological age trajectories or epigenetic age acceleration (EAA).
- Deliver include sample-level beta-value matrices formatted for statistical modeling, group-wise age acceleration summaries, and correlation plots.
- Platform recommendation: methylation arrays for established human clock compatibility; targeted panels for focused marker sets.
Custom clock model development
- Identify age-associated CpGs from genome-wide or reduced-representation methylation data and train a project-specific clock model tailored to your cohort's age range, tissue type, and species.
- Support Elastic Net, Random Forest, and other regression strategies selected based on the number of samples, CpG features, and intended model interpretability.
- Platform recommendation: WGBS (maximum discovery breadth) or RRBS/EM-seq (regulatory-region focus with controlled sequencing budget).
Clock validation and cohort screening with targeted markers
- Once the clock CpG list is defined — whether from published literature, array-based training, or prior discovery — profile those specific loci at high depth across validation or screening cohorts.
- Evaluate cross-sample consistency, per-CpG coverage, missingness rates, and batch structure before integrating clock predictions into the analysis.
- Platform recommendation: targeted bisulfite sequencing panels or multiplex amplicon approaches.
Array-based clock analysis with established models
- Process samples on the Illumina Infinium platform with probe-level QC, functional normalization, and beta-value matrix generation formatted for established clock calculators.
- Compatible with Horvath multi-tissue clock, Hannum clock, GrimAge, DunedinPACE, PhenoAge, and other array-trained models.
- Platform recommendation: Infinium EPIC v2 / 935K methylation microarray.
Low-input and challenging sample types
- Researchers working with FFPE-archived tissues, plasma-derived cfDNA, laser-capture microdissection material, or limited biopsy samples need a platform that tolerates low DNA input and partial fragmentation.
- We assess DNA yield, fragment size distribution, library complexity, and per-CpG coverage risk before committing to a specific assay path, and adjust the analysis plan to account for increased missingness at clock-relevant loci if needed.
- Platform recommendation: RRBS or targeted panels (low-input tolerant); EM-seq for FFPE samples (gentler enzymatic conversion).
Multi-omics aging research
- Integrate DNA methylation clock outputs with transcriptome, whole-genome, proteome, metabolome, or health-phenotype data from the same cohort to build a systems-level picture of aging biology.
- Our Epigenomics Data Analysis Service supports multi-modal integration, biomarker discovery across omics layers, and pathway-level interpretation of age-associated molecular changes.
- Platform recommendation: all methylation platforms are compatible; the integration strategy is defined during the project design phase based on available data types.
Sample Input Requirements and QC Guidelines
Sample quality and input quantity are the most frequently underestimated variables in methylation clock projects. A model trained on 500 ng of high-integrity gDNA will not perform the same way on data from degraded or low-yield material. We evaluate each sample against platform-specific thresholds before library preparation and adjust the clock analysis plan when sample quality falls outside the optimal range.
| Sample Type | Recommended Input Range | Critical QC Parameters | Platform Compatibility Notes |
|---|---|---|---|
| High-quality genomic DNA (fresh or flash-frozen tissue, blood, cultured cells) | WGBS: 500 ng – 1 µg; RRBS/EM-seq: 50–100 ng; targeted panels: 10–50 ng; methylation arrays: 250–500 ng | Concentration (fluorometric), A260/280 (1.8–2.0), A260/230 (≥1.8), high molecular weight integrity by gel or TapeStation | Suitable for all platforms; recommended as the primary sample type for discovery-phase and WGBS-based clock projects |
| FFPE-derived DNA | ≥200 ng when available; feasibility depends on fragment size distribution and amplifiability | Degradation index, bisulfite conversion efficiency, per-locus amplification success rate, missing-CpG risk | EM-seq (enzymatic conversion, gentler than bisulfite) or RRBS recommended; WGBS generally not suitable; analysis pipeline requires coverage-aware filtering and imputation review |
| Cell-free DNA (plasma, serum) | Typically 2–4 mL plasma equivalent; cfDNA input of 1–30 ng post-extraction | cfDNA fragment size distribution (peak at ~166 bp), library complexity, adapter-dimer contamination, end-repair efficiency | Targeted bisulfite panels or ultra-low-input RRBS workflows strongly preferred; whole-genome approaches require considerable optimization and may not achieve sufficient coverage at clock-relevant loci |
| Existing methylation datasets | FASTQ, IDAT, processed beta-value matrix, or per-CpG coverage tables | Metadata completeness, genome build version, array platform version (450K/EPIC/EPICv2), probe annotation compatibility, batch structure, sample-level age and phenotype data | Suitable for reanalysis with alternative clock models, cross-cohort harmonization, meta-analysis, or integration with newly generated data from the same cohort |
End-to-End DNA Methylation Clock Workflow
From project design and sample QC through methylation calling, normalization, and biological age estimation
Primary analysis — Sequence QC, alignment, and data review
- Read-level quality inspection: base quality scores, adapter content, sequence duplication, and methylation-specific bias (bisulfite conversion rate, strand specificity).
- Adapter trimming and low-quality base removal with project-appropriate parameters that preserve methylation information.
- Bisulfite-aware alignment to the selected reference genome (hg38, mm39, or custom non-human reference) using validated aligners. Array data undergoes probe-level QC including detection P-values, bead-count filtering, and control-probe review.
Secondary analysis — Methylation calling, coverage filtering, and feature matrix assembly
- Methylation status determination at each CpG site (and non-CpG context where the clock model requires it).
- Coverage-based CpG filtering: loci below minimum read depth are flagged, and the impact on clock-model completeness is assessed.
- Beta-value matrix generation, genomic annotation (promoters, gene bodies, enhancers, CpG islands, shores, shelves), batch-effect assessment using principal component analysis, and normalization (quantile, BMIQ, or functional normalization depending on platform and cohort structure).
Tertiary clock analysis — Biological age estimation, model training, and interpretation support
We convert normalized methylation profiles into the input format required by the target clock model. For projects using established clocks, we apply published coefficients (Horvath, Hannum, GrimAge, DunedinPACE, PhenoAge, or others) and report predicted age, EAA, and model-specific metrics. For custom clock development, we support feature selection, Elastic Net or Random Forest model training, cross-validation, and independent cohort testing. Outputs include age prediction scatter plots, sample clustering by predicted age, EAA group comparisons, and a written interpretation summarizing the clock results in the context of the study design. Where the project combines methylation data with transcriptome, genome, or phenotype information, our analysis can extend into multi-omics integration.
What Makes Our Clock Service Different
Platform selection driven by the clock question, not by default workflows
We do not route every project through the same methylation platform. WGBS is not automatically the best choice (it may be unnecessarily costly for a focused GrimAge study), and targeted panels are not appropriate for discovery projects. Our scientific team reviews the clock objective, sample type, cohort size, species, and age structure before recommending a platform, ensuring that the data generated is matched to the model strategy.
QC criteria designed for clock reliability, not only for sequencing quality
Standard methylation pipelines assess library complexity, alignment rate, and conversion efficiency. Our pipeline goes further: we evaluate per-CpG coverage at clock-relevant loci, estimate missing-data risk for each sample, quantify batch structure across cohorts, and flag samples whose coverage profile may compromise model performance. These clock-specific QC steps are the difference between a methylation dataset and a clock-ready dataset.
Flexible bioinformatics that supports both established and novel clock methodologies
Whether you need to run a published clock calculator, train a new model from cohort data, harmonize methylation matrices across batches, or combine clock outputs with transcriptomic or genetic data from the same samples, our bioinformatics pipeline is structured to adapt to the analysis plan rather than forcing the data through a fixed reporting template. We also support analysis of existing methylation data from third-party sources, making our service accessible to researchers who have already generated array or sequencing data and need clock-specific processing.
Demo Results: What a Clock Deliverable Package May Include
The examples below illustrate the types of outputs generated during a typical DNA methylation clock project. Actual deliverables vary depending on the platform selected, clock model applied, and scope of bioinformatics support requested.
Demo 1: Methylation beta-value matrix and CpG coverage QC
After alignment and methylation calling, each sample's CpG-level methylation status is compiled into a beta-value matrix. A coverage QC report flags loci with insufficient read depth, conversion-rate anomalies, and sample-level missingness patterns before the data enters clock model input.
This QC step is particularly important for clock projects because missing CpGs at model-relevant loci can shift predicted age estimates even when the rest of the data appears sound.
Demo 2: Biological age prediction and EAA group comparison
After applying the target clock model, predicted biological age is plotted against chronological age for each sample. Group-level epigenetic age acceleration (EAA) is calculated as the residual of biological age regressed on chronological age, and compared across experimental groups.
Outputs include scatter plots with regression lines, EAA box plots by group, and a summary table reporting R², median absolute error, and mean EAA per group.
Demo 3: Custom clock model training and cross-validation
For projects developing custom clocks, model training outputs include Elastic Net or Random Forest cross-validation results, selected CpG feature lists with coefficients or importance scores, training-test correlation scatter plots, and independent cohort validation performance where data is available.
The model coefficients and CpG feature list are documented for potential application to future cohorts.
Frequently Asked Questions
1. Which methylation platform should I choose for my DNA methylation clock project?
Start by asking what you need from the clock: discovery of novel age-associated CpGs, application of an established model, or validation of a defined marker set. WGBS covers the entire methylome and is appropriate when you cannot predict which loci carry age information. RRBS or EM-seq offer efficient coverage of regulatory regions at lower cost. Arrays provide direct compatibility with published human clock models (Horvath, GrimAge, DunedinPACE). Targeted panels deliver the highest cost efficiency when the clock CpG list is already known. We review these factors with you at the project design stage to avoid either over-engineering or under-powering the data.
2. Can CD Genomics process existing methylation data for clock analysis?
Yes. We accept FASTQ files from bisulfite sequencing experiments, IDAT files from Infinium arrays, processed beta-value matrices, or per-CpG coverage tables. The requirements are that the metadata includes genome build, probe or locus annotation, sample grouping information, and chronological age where available. We then apply our clock-specific QC, normalization, and model pipeline to generate biological age estimates and EAA metrics from the existing data.
3. Do you support epigenetic clock analysis in non-human species?
Yes, provided that a reference genome is available for the species. For non-human models, we can align methylation reads to the appropriate reference, call CpG methylation across the genome, and identify age-associated markers. Custom clock model training requires sufficient sample size and age range for statistical power — we assess this during the project design phase. Cross-species clock development (e.g., transferring a human clock to a primate model or building a mouse-specific aging clock) is an area of active methodology support.
4. What happens if important clock CpGs have low or missing coverage in my data?
Our pipeline flags low-coverage and missing clock CpGs in the QC report before model input. Depending on the extent and pattern of missingness, we may recommend coverage-based filtering (removing CpGs with >N% missing across samples), imputation using neighboring CpG correlations, a targeted resequencing panel to rescue the critical loci, or a platform adjustment for subsequent cohorts. The approach is documented in the deliverable report so that the model output can be interpreted with appropriate caution.
5. Can FFPE or cfDNA samples be used for clock studies?
These sample types can be included, but project feasibility must be evaluated on a per-sample basis. Key factors include DNA yield, fragment size distribution, bisulfite conversion efficiency (for FFPE), and library complexity (for cfDNA). We conduct a pre-library feasibility assessment using a small aliquot of the sample material before committing to full-scale processing. For samples that pass the assessment, we typically recommend RRBS, EM-seq, or targeted panels rather than WGBS, as these methods tolerate lower input and partial fragmentation more effectively.
6. What is the difference between biological age and epigenetic age acceleration?
Biological age refers to the predicted age derived from the methylation clock model, which may differ from chronological age. Epigenetic age acceleration (EAA) is the residual of biological age regressed on chronological age — that is, the component of predicted age that cannot be explained by how old the subject is chronologically. Positive EAA indicates that the sample's methylation profile appears biologically older than expected. EAA is the metric most commonly tested against health outcomes, lifestyle factors, environmental exposures, and intervention effects in clock-based studies.
7. What bioinformatics support is available for custom clock model training?
We support Elastic Net regression (the most widely used method in clock development), Random Forest, and other machine-learning approaches depending on the cohort size, number of CpG features, and model interpretability requirements. Our support includes feature selection, training-validation-test split design, cross-validation parameter tuning, model performance evaluation (correlation with age, median absolute error, residuals), and independent cohort validation where data is available. The trained model coefficients are documented for potential use in subsequent projects.
Case Study: DunedinPACE — A Multi-System Methylation Biomarker of the Pace of Aging
Open-access literature case
Background
Belsky and colleagues developed DunedinPACE, a DNA methylation biomarker designed to capture the pace of biological aging from a single blood sample. The study used the Dunedin longitudinal birth cohort, following 1,037 individuals born in 1972–1973 in Dunedin, New Zealand, with repeated physiological measurements across four time points spanning two decades (ages 26, 32, 38, and 45 years). The Pace of Aging phenotype was calculated as the rate of decline across 19 indicators of organ-system integrity, covering the cardiovascular, metabolic, renal, immune, dental, pulmonary, and cognitive systems.
Methods
The study used Elastic Net regression on Illumina 450K and EPIC array data to select CpG sites whose combined methylation signature most strongly predicted the multi-system Pace of Aging. Model training and validation were conducted within the Dunedin cohort using repeated measures across four time points, and the final model was tested in five independent validation cohorts from the US and Europe.
Results
The resulting DunedinPACE measure selected 173 CpG sites and showed high test-retest reliability (intraclass correlation > 0.9). It was associated with morbidity, disability, and mortality in independent validation cohorts, and added incremental prediction beyond the GrimAge clock for incident disease and mortality outcomes. DunedinPACE also detected faster biological aging in young adults who had experienced childhood adversity, demonstrating sensitivity to early-life social determinants of aging.
Figure 1 from Belsky et al., eLife 2022 (CC BY 4.0).
Significance for clock research projects
For researchers planning clock studies in their own cohorts, the DunedinPACE study illustrates two principles that directly affect service project design. First, the clock is only as robust as the phenotypic data: training against longitudinal multi-system decline produced a biomarker with functional interpretability that training against chronological age alone would not have captured. Second, the methylation platform (here, the Infinium 450K/EPIC array) must be selected with the model strategy in mind — DunedinPACE coefficients are array-specific and cannot be directly transferred to WGBS or targeted panel data without careful cross-platform normalization and validation.
DunedinPACE, a DNA methylation biomarker of the pace of aging.
References
- Horvath S. DNA methylation age of human tissues and cell types. Genome Biology. 2013;14:R115.
- Lu AT, Quach A, Wilson JG, Reiner AP, Aviv A, Raj K, et al. DNA methylation GrimAge strongly predicts lifespan and healthspan. Aging. 2019;11:303–327.
- Belsky DW, Caspi A, Corcoran DL, Sugden K, Poulton R, Arseneault L, et al. DunedinPACE, a DNA methylation biomarker of the pace of aging. eLife. 2022;11:e73420.
- Bell CG, Lowe R, Adams PD, Baccarelli AA, Beck S, Bell JT, et al. DNA methylation aging clocks: challenges and recommendations. Genome Biology. 2019;20:249.
