Inter-laboratory adaption of age estimation models by DNA methylation analysis-problems and solutions

In recent years, a lot of age prediction models based on different CpG motives in different cell types were published determining the biological age of a person by DNA methylation. For a general employment of this technique, maybe even as a routine method, the cross-laboratory application of such models has to be examined. Therefore, we tested two different published age prediction models for blood and mouth swab samples with regard to prediction accuracy (Bekaert et al Epigenetics 10:922-930, 2015a; Bekaert et al Forensic Sci Int Genet Suppl Ser 5:e144-e145, 2015b). Both models are based on CpG sites of four genes (ASPA, EDARADD, PDE4-C, and ELOVL2), but with a different combination of CpGs for the two tissue types. A mean absolute difference (MAD) between chronological and predicted age of 9.84 and 8.32 years for blood and buccal swab models could be demonstrated, respectively, which is significantly worse than the published data, probably due to higher DNA methylation variances in some CpGs. By retraining both prediction models, the prediction accuracy could be improved to a MAD of 5.55 and 4.65 years for the renewed blood and buccal swab model, respectively. This study demonstrates the usefulness of effective DNA standards to normalize DNA methylation data for better comparison of study results.