Article ID: 867
Age estimation based on the methylation rates of DNA in blood samples can provide useful information for narrowing down candidates of unidentified donors and might be helpful in supporting criminal investigations. In our previous study, we showed that Bekaert's age-prediction model using pyrosequencing can estimate the age of Japanese blood samples with high accuracy. Recently, Garali et al. reported a new age-prediction model that requires a smaller amount of DNA compared to Bekaert's model, using the same platform. In the present study, we validated whether Garali's age-prediction model can be applied to Japanese blood samples. In addition, we verified whether these prediction models can be applied to aged bloodstain samples, which are common in criminal investigations. The accuracy of Garali's model on Japanese blood samples (n=143) was 3.19 years in terms of absolute mean error (MAE) and 4.24 in terms of root mean square error (RMSE), which were comparable to those of Bekaert's model (MAE 3.24 and RMSE 4.12). Although Garali's model alone could predict age with high accuracy, it was suggested that using both models for age prediction is preferrable when DNA yields are sufficient because Bekaert's model could probabilistically reduce prediction error due to marker multiplicity. At nearly all CpG sites examined, no significant differences were observed in methylation rates between aged bloodstain samples (stored for 1 week to 14 years) and blood controls. Similarly, differences in predicted ages between Japanese blood and aged bloodstain samples were very small in both models. These results suggest that both age-prediction models can be applied to aged bloodstain samples with the same high accuracy as when applied to fresh blood, even when samples have been stored for a long period of time.
