Related Experiment Video
Updated: Nov 21, 2025
14:56
Sample Preparation to Bioinformatics Analysis of DNA Methylation: Association Strategy for Obesity and Related Trait Studies
Published on: May 6, 2022
4.9K
Artificial neural network, predictor variables and sensitivity threshold for DNA methylation-based age prediction
Zhonghui Thong1, Jolena Ying Ying Tan2, Eileen Shuzhen Loo2
1DNA Profiling Laboratory, Biology Division, Health Sciences Authority, 11 Outram Road, Singapore, 169078, Singapore. thong_zhonghui@hsa.gov.sg.
Scientific Reports
|January 19, 2021
Summary
Artificial neural networks (ANNs) provide more accurate epigenetic age prediction than regression models. This study validates ANNs for forensic use, showing minimal impact from ethnicity or sex, and enabling prediction with reduced DNA input.
Area of Science:
- Forensic epigenetics
- Biomarker discovery
Background:
- Epigenetic age prediction using DNA methylation is crucial for forensics.
- Previous models lacked validation for ethnicity, sex, and minimal DNA input.
Purpose of the Study:
- To compare regression and artificial neural network (ANN) models for epigenetic age prediction.
- To investigate the influence of ethnicity and sex on age prediction accuracy.
- To determine the minimum DNA input required for reliable age prediction using pyrosequencing.
Main Methods:
- Evaluated regression and ANN models on 333 blood samples from diverse ethnicities.
- Assessed the impact of ethnicity and sex on age prediction accuracy.
- Developed and tested a single-locus ANN model with reduced DNA input (25 ng).
Main Results:
- ANN models demonstrated higher accuracy in age prediction compared to regression models.
- Ethnicity did not significantly impact age prediction accuracy across Chinese, Malay, and Indian populations.
- Sex had a marginal effect (overestimation in males) but did not compromise overall prediction accuracy.
- A one-locus, dual CpG model using 25 ng DNA proved sufficient for forensic age prediction.
Conclusions:
- ANNs are superior to regression models for epigenetic age prediction in forensic contexts.
- Validated ANN models are reliable across diverse ethnicities and minimally affected by sex.
- Developed a low-input DNA method (25 ng) for forensic age estimation, enhancing its practical applicability.
