A central part of forensic genetic case work is comparison between reference and traces STR-profiles, an often manual and time-consuming process involving visual inspection of electropherograms and DNA-profile tables. Match-matrices highlighting full inclusions or near matches may aid case workers especially in complex cases where many comparisons need review. Although they can be analytically simplistic and have critical shortcomings, such as finding true donors with weak contributions to mixtures, as well as limited case overview, particularly in cases with multiple reference and trace profiles.

We present an analytical and visual approach that enables inference of relationships between traces and reference profiles, drawing from both multivariate and structure analysis widely used in traditional population genetics, but without assuming any specific mathematical and biological properties. Here we employ Principal Coordinate Analysis (PCoA) and Non-negative Matrix Factorization (NMF) on real case data consisting of reference profiles as well as single-donor and mixture profiles from traces. The PCoA ordinates multidimensional STR-data across loci to display profile relationships based on matrix similarities/dissimilarities. NMF is used as an unsupervised machine learning technique to assign profiles to genetic groups, thereby potentially identifying genetically admixed trace samples of known and unknown contributors. Trace and reference samples were genotyped using the Globalfiler IQC PCR amplification kit, and the input data consists of allele identity and per locus normalized peak heights to account for contributor proportions.

We show that the PCoA and NMF in concert can i) assist the case worker in obtaining similarity overview across both reference and trace profiles, ii) distinguish potential associations between reference profiles and mixture trace profile with variable contributions (i.e. from full inclusion to multiple allele drop-outs) and iii) potentially highlight profiles with unknown contributors that may be derived for investigative purposes (e.g. deduction of assumed profile from mixtures using probabilistic genotyping).

Authors

• Andreas Kelager (Section of Forensic Genetics, Department of Forensic Medicine, University of Copenhagen, Denmark)
• Aleksandra Laura Pach (Section of Forensic Genetics, Department of Forensic Medicine, University of Copenhagen, Denmark)
• Anders Buchard (Section of Forensic Genetics, Department of Forensic Medicine, University of Copenhagen, Denmark)