Accurate estimation of chronological age in adolescents and young adults is a key challenge in forensics, particularly in legal contexts where individuals lack documentation and age assessment has direct consequences for rights and protection. DNA methylation-based methods are among the most promising biological approaches; however, most existing epigenetic clocks are not optimized for this developmental period.

We present a two-stage investigation of blood-based DNA methylation age prediction in individuals aged 12–25 years. First, we developed the Predictor for Adolescents and Young Adults (PAYA), based on 267 CpG sites from blood samples. PAYA demonstrated high accuracy in independent validation data, achieving a median absolute deviation (MAD) of approximately 0.7 years in 18-year-old individuals. However, further analyses revealed systematic bias across the broader 12–25 age range.

To address these limitations, we evaluated multiple existing predictors across a multi-cohort dataset comprising 926 samples generated using Illumina 450K, EPIC, and EPIC v2 arrays. We assessed robustness across preprocessing strategies, technical variation, and cohort heterogeneity, and tested different approaches for optimizing age estimation. Based on these analyses, we developed Ensaya, an ensemble model combining top-performing predictors. Ensaya achieved improved overall performance across the full age range, with a MAD of approximately 0.8 years for individuals aged 12–25, alongside reduced age-dependent bias, supporting the importance of age-specific and ensemble-based modeling for robust and legally accurate age estimation in adolescents and young adults.

Authors

• Kristina Stenløkk (Oslo Univeristy Hospital, Norway)
• Øyvind Bleka (Oslo Univeristy Hospital, Norway)
• Pål Skage Dahlberg (Oslo Univeristy Hospital, Norway)
• Veslemøy Rolseth (Oslo Univeristy Hospital, Norway)