AgingClockBench

Benchmark biological aging clocks on your data in minutes.

Multiple biological aging clocks exist — PhenoAge, KDM, DunedinPACE — but no standard tool lets researchers compare them side-by-side. AgingClockBench is the first open-source Python package implementing multiple clocks with a unified interface and reproducible mortality-validated benchmarking.

📖 Documentation & Examples

---

Install

pip install agingclockbench

Requires Python 3.11+.

---

Quick Start

from agingclockbench import PhenoAge, KDM, BenchmarkSuite
from agingclockbench.datasets import load_nhanes_sample

# Load bundled NHANES 1999-2000 (N=4,086, 20-year mortality follow-up)
df = load_nhanes_sample()

# Compute biological ages
results = {
    "PhenoAge": PhenoAge().transform(df),
    "KDM":      KDM().transform(df),
}

# Benchmark against mortality
suite = BenchmarkSuite(mortality_col="mortstat", followup_col="permth_exm")
report = suite.run(df, results)

print(report.to_dataframe())

Clock     Pearson r  Mort HR (per SD accel)  Mort p-value
PhenoAge      0.930                    1.83      0.000001
KDM           0.677                    1.41      0.000001

report.plot_km_survival()      # Kaplan-Meier by acceleration quartile
report.plot_comparison()       # biological age vs chronological age
report.to_html("report.html")  # interactive Plotly report

---

CLI

# Benchmark on bundled NHANES with HTML report
agingclockbench benchmark --data bundled --clocks all --report

# Your own data
agingclockbench benchmark \
  --data my_cohort.csv \
  --clocks PhenoAge KDM \
  --mortality-col vital_status \
  --followup-col followup_months \
  --output ./results/

---

Implemented Clocks

Clock	Reference	Biomarkers	Key metric (NHANES)
PhenoAge	Levine et al. 2018 Aging Cell	9 blood	Pearson r=0.93, HR=1.83
KDM	Klemera & Doubal 2006 Mech Ageing Dev	8 blood	Pearson r=0.68, HR=1.41
DunedinPACEProxy	Proxy (NOT Belsky 2022)	7 blood	pace corr w/ PhenoAge accel r=0.84

Note: DunedinPACEProxy is a blood-biomarker approximation. The real DunedinPACE requires DNA methylation data (Illumina EPIC array).

---

Features

• ✅ Unified interface — all clocks share the same transform() API
• ✅ Validated — PhenoAge implementation cross-validated against reference; zero numerical difference on N=4,086 NHANES participants
• ✅ Mortality benchmarking — Cox PH hazard ratios, Kaplan-Meier curves
• ✅ Bundled data — NHANES 1999-2000 with 20-year mortality follow-up, ready to use
• ✅ Interactive reports — Plotly HTML with comparison plots and benchmark table
• ✅ CLI tool — agingclockbench benchmark works out of the box
• ✅ 89% test coverage — 91 tests, CI/CD on every push

---

FAQ

How do I add a new clock?

Implement the BaseClock interface in a new file under src/agingclockbench/clocks/. See CONTRIBUTING.md for step-by-step instructions and BaseClock source.

What's the difference between PhenoAge, KDM, and DunedinPACEProxy?

Clock	Biomarkers	Key Metric	Use Case
PhenoAge	9 blood	HR=1.83 per 10yr accel	Best mortality prediction; recommended for research
KDM	8 blood	HR=1.41 per 10yr accel	Classical approach; simpler model
DunedinPACEProxy	7 blood	r=0.84 w/ PhenoAge	Blood-based proxy (real DunedinPACE requires DNA methylation)

Can I use my own data?

Yes! Pass any pandas DataFrame with the required biomarker columns:

import pandas as pd
from agingclockbench import PhenoAge

df = pd.read_csv("my_cohort.csv")
result = PhenoAge().transform(df)
print(result.biological_ages)

How do I interpret the HTML report?

The report generated by report.to_html("report.html") contains three sections:

1. Benchmark Table — Pearson r (correlation), Spearman r, Mortality HR (hazard ratio per 10-year acceleration), and p-value
2. Bland-Altman Plot — Shows mean bias and ±95% limits of agreement; tight limits indicate consistent predictions
3. Kaplan-Meier Survival Curves — Stratified by clock acceleration quartile; steeper curves indicate stronger mortality association

What if I get a "missing biomarker" error?

All 9 biomarkers are required for PhenoAge; 8 for KDM. Check that your DataFrame contains:

• For PhenoAge: albumin_g_l, creatinine_umol_l, glucose_mmol_l, crp_mg_l, lymphocyte_pct, mcv_fl, rdw_pct, alkaline_phosphatase_u_l, wbc_10k_ul
• For KDM: All of the above except crp_mg_l

Drop rows with missing values before calling .transform().

---

Documentation & Learning

Getting started: See Quick Start above for a minimal working example.

API reference:

from agingclockbench import PhenoAge, KDM, BenchmarkSuite
help(PhenoAge)        # View docstring and parameters
help(BenchmarkSuite)  # View benchmarking options

Adding a new clock: See CONTRIBUTING.md for step-by-step instructions.

Example notebooks: See examples/ for Jupyter notebooks.

---

Citation

If you use AgingClockBench in your research, please cite:

@software{geddam2026agingclockbench,
  author = {Geddam, Aaditya},
  title  = {AgingClockBench: Benchmarking biological aging clocks},
  url    = {https://github.com/aadityageddam-ux/aging_clock_bench},
  year   = {2026}
}

Also cite the underlying clock papers:

• Levine ME, et al. Aging Cell. 2018. (PhenoAge)
• Klemera P, Doubal S. Mech Ageing Dev. 2006. (KDM)

---

Contributing

Contributions welcome! See CONTRIBUTING.md.

To add a new clock, implement the BaseClock interface — see the FAQ section above.

---

License

MIT © Aaditya Geddam