Calibrating epigenetic clocks with training data error.

Area of Science:

• Wildlife biology
• Molecular ecology
• Biotechnology

Background:

• Accurate animal age data are vital for wildlife population management.
• Epigenetic clocks offer a molecular method for age prediction using DNA methylation.
• Developing epigenetic clocks is hindered by the lack of accurately aged calibration samples for most species.

Purpose of the Study:

• To determine the tolerance threshold for age error in epigenetic clock training data.
• To assess the impact of inaccurate calibration data on age prediction accuracy.
• To inform the development and application of epigenetic clocks in wildlife research.

Main Methods:

• Utilized four public datasets for epigenetic clock development.
• Artificially introduced incremental age errors (1%) into training data.
• Validated model performance against an independent set of known ages.

Main Results:

• A statistically significant increase in age prediction error (Cohen's d >0.2) was observed when training data age error surpassed 22%.
• The effect size of age prediction error increased linearly with the error in training data.
• This error tolerance threshold was found to be independent of sample size.

Conclusions:

• Epigenetic clock development requires high-quality, accurately aged calibration data for precise age prediction.
• The acceptable level of error in training data depends on the downstream application's precision requirements.
• For applications requiring relative age order, less accurate calibration data may suffice, but precise age estimation may be unattainable with flawed data.