Age prediction of contributors to two-person DNA mixtures via deconvolution of autosomal DNA methylation profiles

DNA methylation (DNAm) profiling has proved to be a reliable method for estimating the chronological age of an unknown sample donor, offering valuable leads in police investigations. However, a major limitation of current approaches using autosomal age-predictive CpGs is their inapplicability to multi-donor DNA samples (DNA mixtures), which are frequently encountered in forensic casework. Here, we present a novel approach, UnMixMe (UnMix DNA Methylation profiles), to deconvolute the DNAm profile of an unknown individual (the suspect) from a two-person DNA mixture and subsequently use this profile to predict the suspect's chronological age. Importantly, it relies on knowing 1) the DNAm profile of the other contributor (the victim) and 2) the DNA mixture ratio between the two contributors. The latter was estimated via the traditional STR profile for human identification. We tested our approach on mock trace DNA mixtures prepared from blood samples with varying suspect-to-victim ratios (10:1, 4:1, 2:1, 1:1, 1:2, 1:4, and 1:10) from two male-female pairs. We measured DNAm levels (β-values) using the Illumina EPIC v2.0 microarray and deconvoluted the DNAm profiles of the DNA mixtures to retrieve the DNAm profile of the suspect. The age of the suspect was predicted using four array-based epigenetic clocks (BLUP, EN, Horvath, and skinHorvath clock). We achieved age prediction accuracy comparable to that of the single-source suspect DNA sample, except for DNA mixtures with a low suspect-to-victim ratio (1:4 and 1:10). We identified three main factors affecting the age prediction accuracy: 1) precision of the DNAm technology, 2) accuracy of DNA mixture ratio estimation, and 3) accuracy of the prediction model used. Importantly, the age difference between DNA mixture contributors did not influence prediction accuracy. With this proof-of-concept study, we establish that autosomal DNAm profiles from two-person DNA mixtures can be successfully deconvoluted when one contributor is known and highlight the potential of this method for predicting chronological age in mixed DNA samples.