DNA mixtures interpretation – A proof-of-concept multi-software comparison highlighting different probabilistic methods’ performances on challenging samples

Abstract

The present study investigated the capabilities and performances of semi-continuous and fully-continuous probabilistic approaches to DNA mixtures interpretation, particularly when dealing with Low-Template DNA mixtures. Five statistical interpretation software, such as Lab Retriever and LRmix Studio - involving semi-continuous algorithms - and DNA•VIEW, EuroForMix and STRmix- employing fully-continuous formulae - were employed to calculate likelihood ratio, comparing the prosecution and the defense hypotheses relative to a series of on-purpose prepared DNA mixtures that respectively contained 2 and 3 known contributors. National Institute of Standards and Technologies (NIST) certified templates were used for samples set up, which contained different DNA amounts for each contributor. 2-person mixtures have been prepared with proportions equal to 1:1, 19:1 and 1:19 in terms of DNA concentration. Conversely, three person mixtures were constituted by proportions equal to 20:9:1, 8:1:1, 6:3:1 and 1:1:1 in terms of DNA concentration. Furthermore, 8 equally-proportioned 3-person mixtures were prepared by means of scalar dilutions starting from an overall amount of 0.500 ng, then ranging up to DNA samples with concentrations equal to 0.004 ng (i.e. Low-Template DNA). DNA mixtures were set up in triplicate and amplified with 7 DNA amplification kits (i.e. GlobalFiler PCR Amplification Kit, NGM SElect PCR Amplification Kit, MiniFiler PCR Amplification Kit, Power Plex Fusion, PowerPlex 6C Matrix System, Power Plex ESI 17 Fast and Power Plex ESX 17 Fast) in order to evaluate whether the selection of a certain kit might represent a bias factor, capable of altering the whole interpretation process. A multi-software approach helped us to highlight any trend in the likelihood ratio results provided by semi- and fully-continuous software. As a matter of fact, fully-continuous computations provided different (higher) results in terms of degrees of magnitude of the likelihood ratio values with respect to those from the semi-continuous approach, regardless of the amplification kit that was utilized.

Four model variants within a continuous forensic DNA mixture interpretation framework: Effects on evidential inference and reporting

“…It has also been demonstrated that factors such as PCR and the content of allele frequency databases have an impact on the variation in the LR computed using a continuous method [ 28 , 29 ]. Recent work has demonstrated that differences in output between semi-continuous and continuous systems result in clear differences in the LRs for some samples [ 10 , 30 , 31 ]. Despite these studies, comparisons between continuous probabilistic systems are not readily available in the literature, though some examples using small datasets do exist.…”

“…For example, Morimoto et al [ 31 ] compare the continuous system Kongoh to another continuous system, EuroForMix, and demonstrate that for most high-template simple mixtures tested the LR outcomes were similar; however differences in LRs obtained from each model were obtained for more complex mixtures wherein the authors attributed the variation in outputs as “differences in the computational principle of estimating peak height variances”. Though reports of inter-model comparisons in the scientific literature do exist, sometimes resulting in the use of multiple softwares to test one item of evidence [ 30 ], the published work use limited datasets or do not replicate the runs; thus, in this work, and pursuant to PCAST’s recommendation to publish large-scale studies, we supplement the forensic and scientific literary record by examining the variability between results obtained from four variants of CEESIt [ 32 ], a tool that computes a continuous LR for a person of interest. In addition to the LR, CEESIt computes the LR distribution by random sampling of genotypes conditioned on the defense’s hypothesis, as well as the so-called p- value for the LR, which is the proportion of LRs sampled that are at least as large as the LR for the person of interest.…”

Interpretation continues to be the main weakness in criminal justice systems: Developing roles of the expert witness and court

“…(Bleka, Benschop, Storvik, & Gill, ) compared a qualitative versus quantitative mixture models, showing that the latter performs more efficiently because they take account of peak height information. (Alladio et al, ) compared three quantitative models (STRmix, EuroFormMix and DNAview) showing consistency of results, that is, different models built using different assumptions returned similar results when tested against the same data. Qualitative models, Lab Retriever and LRmix Studio, behaved consistently, but tended to give lower LRs compared to quantitative models. Different models/databases may be used to solve the same problem.…”

CaseSolver: An investigative open source expert system based on EuroForMix

“…3 and this is covered by separate validation studies reported elsewhere (e.g. [26,35,36]). Supplement D describes program checking carried out using R scripts that are available at the webpage www.euroformix/evaluations in a file named 'NumEval 2.1.0'.…”