Inter-laboratory variation in interpretations, comparisons, and statistical analyses of DNA mixtures

DNAmix2021 was a large-scale inter-laboratory study conducted to assess variation in interpretations, comparisons, and statistical analyses of DNA mixtures. Analyses were based on 765 responses by 106 participants from 52 labs. Eight distinct DNA mixtures were created, each of which was provided to participants as a contributor packet (the person of interest (POI) was in the mixture), or as a non-contributor packet (the POI was NOT in the mixture). Accuracy on contributor packets was notably associated with the percent of DNA contributed by the POI: packets in which the POI contributed less than 8% of the DNA (≤25 pg) had much higher rates of false exclusions (contrary to ground truth) and indeterminate responses, compared to packets in which the POI contributed more DNA. A lab's ability to discern a low-level contributor is largely a function of its operating procedures: the community may wish to consider whether the exclusion of very low-level contributors can or should be considered incorrect. Most false inclusions were reported on one non-contributor packet that had high allele sharing with a POI that was a sibling of a contributor to the mixture. Most false exclusions and false inclusions were associated with incorrect estimates of the number of contributors. The few false inclusions may also be explained as a combination of inclusions without supporting statistics and/or not conditioning on reference profiles. The only likelihood ratios indicating very strong support that were contrary to ground truth were on contributor packets with a low proportion of the DNA contributed by the POI.