Twin-pred: a method to distinguish monozygotic twins in forensic science application

Monozygotic (MZ) twins discrimination in forensic science remains an unsolved point. Nowadays, conventional DNA profiling techniques use the analysis of the Short tandem repeats (STR) to distinguish between suspects. However, since MZ twins share the same DNA sequences, their discrimination using STR analysis presents several limitations. To overcome these limitations, scientists focused their attention on the study of DNA epigenetic modifications, and in particular on DNA methylation. DNA methylation is an epigenetic DNA modification that occurs at the 5′ positions of cytosine in CpG dinucleotides. So far, many works have identified epigenetics as a possible suitable solution for identical twins discrimination (Marqueta-Gracia et al., 2018; Vidaki et al., 2017b). In our study, we set up a suitable protocol to distinguish between identical twins in forensic cases. We identified an end-to-end approach, which ranged from DNA extraction to final statistical analysis. To do that, we first set up the experiments as an analogy of a forensic case experiment. As starting material, we used the buccal swabs, often used in forensic science research. In detail, we collected from two couple of MZ volunteers' twins buccal swab samples, and we then extracted the total DNA. We then prepared NGS libraries using bisulfited conversion strategies, considered the gold standard in methylation study, to potentially target every single methylated cytosine state present in the twins genomes. To better asses a forensic case situation, for each couple we generated a set of Reference experiment libraries and four different Test libraries, two per individual. In detail, Reference experiment libraries were created from sufficient quantities of DNA, simulating a standard buccal swab sampling while the Test libraries were created from a small quantity of starting DNA, simulating the limiting DNA quantity that may be available from a crime scene. Both reference and test data were analyzed using bioinformatic tools specific for DNA methylated samples (Bismark Alignment software and Methylkit). Once all cytosine methylation states were decided, we set up a statistical approach to infer the probability of one test sample being either one of the reference samples. Briefly, the binomial probability of each informative sequenced CpG was used as a part to derive a discriminatory estimate for each twin based on the known level of methylation at that site in the Reference data. The aggregation of such probability components for all sites supplied the final association to the twin. We permuted several analysis parameters to find the best set and we assessed the reliability of the prediction by performing bootstrap analysis on the sets of parameters that gave back the best accuracy in the calling. Lastly, we studied the trend of prediction accuracy of every single parameter passed into the function, to screen how and if they could affect the prediction.