Evaluation of Inter-laboratory Probabilistic Genotyping Parameters

Wednesday November 2nd, 2022 // 9:00 am - 9:35 am // Woodrow Wilson Ballroom

The authors describe an interlaboratory DNA interpretation experiment using data from eight forensic biology laboratories and the probabilistic genotyping system STRmix™. Each laboratory contributed twenty mixtures of two to four contributors from their validation studies along with their STRmix™ parameters including STRmix™ kit files and stutters. The extent to which STRmix™ parameters which have been optimized by individual laboratories can be applied to mixture data created in other laboratories was investigated. Previous research concentrated solely on sensitivity studies involving a single type of PCR kit and a small number of laboratories. In this study, we looked at laboratories that used different PCR kits, different numbers of PCR cycles, and different versions of STRmix™. All eight laboratories’ STRmix™ kits were updated to work on STRmix™ version 2.9.0 and all mixtures were evaluated with this version. This interlaboratory study using known sets of donors is a study of the typed called for by the NIST foundation review of DNA Mixture Interpretation.


All mixtures were deconvoluted using the same sets of propositions and compared to a database of 10,000 random profiles seeded with the 81 known DNA donors used to make the various validation mixtures. The true donor LR and the rank of the LR in the database of 10,081 profiles were recorded, along with the largest non-donor LR and the rank of the first LR=0 for non-donors. All two person mixtures were deconvoluted with three sets of denominator propositions, the three person mixtures used seven different deconvolutions, and the four person mixtures had five different deconvolution strategies. The deconvolutions ranged from using no conditioning profiles to deconvolutions that were fully conditioned for each donor in turn.


Six laboratories used GlobalFiler™ and two laboratories used Investigator® 24Plex QS. Three laboratories used 28 PCR cycles with the remaining using 29 cycles. Each STR kit had both 28 and 29 cycle data. Half of the laboratories only considered standard stutters (one repeat forward and one repeat back stutters) while the other half added in non-standard stutters consisting of double back stutter and half-repeat back stutter. All STRmix™ inputs consisted of the input files provided by the participants, with analytical thresholds (AT) ranging from 40 to 200 rfu, with an even mix of color-specific and across-the-board ATs.


A total of 6119 STRmix™ interpretations were carried out and roughly 61,000,000 LRs assigned. All 2-, 3-, and 4-person mixture likelihood ratios were similar across all 8 laboratories, using both the cognate STRmix™ parameters (LabA STRmix™ on LabA mixtures) as well as the non-cognate parameters (LabB-LabH on LabA mixtures). In addition, the rank of the true donors against the total database of 10081 profiles showed good consistency between the eight STRmix™ kits. This ranking was used as the primary metric of STRmix™ performance. STRmix™ is a very robust probabilistic genotyping system as shown in this study. In addition, this study can be a proof of concept for the direct exchange of mixture data between laboratories that use STRmix™, particularly if there is an investigative reason to do so. It is hoped that this study will be considered a contribution in the efforts to establish the reliability of using STRmix™ in casework as called for by NIST.


Tim Kalafut

Associate Professor of Forensic Science, Sam Houston State University

After over twenty years as a practicing forensic DNA analyst, most with the US Army Crime Lab, Tim Kalafut made a mid-life career change into the world of academia. His new role allows him to have freedom to take on new projects, and hopefully shorten the learning curve for the next generation of forensic scientists.

Submit Questions