STRmix™ Collaborative Exercise on DNA Mixture Interpretation
Thursday September 26th, 2019 // 9:00 am - 9:30 am // Oasis 1-2
Reproducibility is one of the main principles of the scientific method. Reproducibility in forensic DNA interpretation typically relates to the reported match statistic. NIST have undertaken a number of inter-laboratory studies to assess reproducibility and variability in DNA profile analysis, most recently with MIX13 (Butler et. al, 2018).
The development of probabilistic genotyping (PG) solutions using semi-continuous and continuous models was partly in a response to intra and inter laboratory differences to improve reproducibility of the results. However, there are still some aspects of DNA profile interpretation that are not controlled by PG software, and variability is to be expected.
More recently a number of inter-laboratory studies conducted with the use PG solutions have been reported such as the use of LRmix and LRmix Studio in the Euroforgen-NoE (L.Prieto et. al, 2014) and GHEP-ISFG (P.A.Barrio et. al, 2018) collaborative exercises, and the use of STRmix™ in the study by Stuart Cooper et. al (2015). Lessons can be learnt from each of these studies; however, there is still a general concern of the limited reproducibility in these studies. This has resulted in discussions for standardisation or guidelines within, and between, forensic laboratories. However, this standardisation can only be achieved after we identify the sources of variation.
In this collaborative exercise of 174 participants, across 42 laboratories, we aim to define the sources of variation. Two complex mixtures from the PROVEDIt dataset (L.E.Alfonse et. al, 2018) were analysed by each participant and propositions based on the provided case circumstances were determined by each participant. Allele Frequency, Fst, and PG parameters were fixed.
We show the reproducibility and repeatability from the submitted results, with the likelihood ratio ranging from 2.02 × 104 to 7.92 × 106 (Sample 1) and 2.21 × 1028 to 2.43 × 1029 (Sample 2). Differences in the LRs calculated for each submission can be attributed to (from largest to smallest effect):
- Varying the number of contributors assumed when interpreting a profile
- Exclusion of some loci when interpreting a profile
- Differences in CE data analysis methods, leading to variation in peak heights
- Run-to-run variation due to random sampling inherent to the MCMC method
We show a convergence of results within, and between, laboratories. The reproducibility of LR reported in this exercise can be attributed to the use of PG software and supports the ongoing transition of forensic laboratories to probabilistic genotyping.
Scientist Developer, Institute of Environmental and Science Research (ESR)
Kevin Cheng is a Scientist Developer at the Institute of Environmental and Science Research (ESR). He is involved with developmental research and validation for probabilistic genotyping software, and teaching the concepts of probabilistic genotyping when required, most recently in Beijing.Submit Questions