DNAメチル化率を指標としたBekaertの年齢推定法の日本人血液試料による検証と新規予測モデルの構築

抄録

 Age estimation by use of DNA methylation rate have recently been one of the hottest topics in forensic genetics, thus many age prediction models have been developed. The age prediction model developed by Bekaert et al was evaluated as the best method comparing six blood-based models in 2019. In this study, Bekaert's age estimation model was validated with 143 Japanese blood samples (98 living and 45 dead blood samples). The difference between duplicate measurement of DNA methylation rate with pyrosequencer was less than one-year per one-gene in age prediction result. Overall mean absolute error (MAE) was 3.24, and root mean square error (RMSE) was 4.13. MAE of the elder group was larger than that of the young group. No statistically significant difference in absolute prediction error was observed between living and deceased or male and female blood samples. Furthermore, a novel model was developed with support vector regression. The MAE and RMSE were 3.49 and 4.51 for test set, respectively. In total, 200 ng of pre-bisulfite-treated DNA was essential to perform these predictions. To reduce the required amount of DNA, new models based on only ELOVL2 methylation rate requiring only 40 ng of DNA were also developed. The MAE and RMSE were 3.15 and 4.20 for test set, respectively. These results suggest that both Bekaert's age estimation model and our novel model have great potential in estimating one's age on forensic practices with some points to note.