Back to Presentations
Casework Validation of Genetic Calculator Mixture Interpretation
M.W. Perlin and B.W. Duceman, "Casework validation of genetic calculator mixture interpretation", American Academy of Forensic Sciences 62nd Annual Meeting, Seattle, WA, 25-Feb-2010.
This talk describes a validation study of TrueAllele® computer interpretation of DNA mixture evidence based on match information. We observe that, over sixteen mixture data items, the computer was a million times more informative than human review without a victim reference, and 50 thousand times more informative when using one. In addition to establishing efficacy, we also measure the reproducibility of computer interpretation. We note that human and computer review use the same principles, but differ in how they examine data -- the computer uses quantitative peak heights ("a better fit's more likely it"), while human inclusion methods view the data qualitatively ("every pair gets equal share"). We conclude that TrueAllele provides a reliable method for DNA mixture interpretation.
After attending this presentation, attendees will better understand how to conduct a DNA mixture validation study, how to measure the efficacy and reproducibility of any DNA interpretation method, and why computer interpretation of DNA evidence can be more informative than manual review.
The presentation will impact the forensic community by enabling practitioners to conduct DNA mixture validation studies on interpretation methods that they would like to present in court.
Interpreting DNA mixtures can be challenging. With the advent of statistical computing, one can reproducibly infer consistent, highly informative results. Such reliable mixture inference is critical for the admissibility of scientific evidence. This paper establishes the efficacy of computer-based genetic calculator mixture interpretation by comparing inferred match information on adjudicated mixture cases relative to currently used manual methods. It also demonstrates the reproducibility of the computer's results.
The key mixture interpretation task is inferring a questioned genotype of an unknown contributor. When there is uncertainty in an inferred genotype, allele pairs are assigned a probability distribution that describes this uncertainty. Different mixture interpretation methods may infer different genotype distributions.
A genetic calculator provides a statistical computer approach that infers genotypes by hypothesizing all feasible solutions, comparing these with observed STR peak height data, and assigning higher probabilities to genotype hypotheses that better fit the data. We examined two quantitative inference methods:
- TA1, which uses a known victim genotype to help infer the other unknown contributor, and
- TA2 that does not use a victim genotype, but instead infers two unknown contributors.
There are also qualitative list-based inclusion methods that apply peak thresholds:
- CLR, which uses a known victim genotype, and
- CPI, a qualitative approach that does not use a victim genotype.
The Likelihood Ratio (LR) is the generally accepted forensic science measure of match rarity. The LR gives the probability of a match between the evidence genotype and a suspect, relative to a match with a random person. The data-inferred evidence genotypes above (TA1, TA2, CLR, CPI) each produce a LR match statistic when their probability distribution is substituted into a generic LR match formula.
The efficacy of the genetic calculator was determined by comparing its LR match information to other methods. In particular, the LR logarithm (i.e., order of magnitude, or powers of ten) was determined on eight adjudicated cases for the two unknown TA2 computer method, and compared with that of the reported CPI value. Whereas the average log(LR) information for CPI was 7 (LR = 10 million to one), the average match information on these same cases with TA2 was 13 (LR = 10 trillion). This shows a six order of magnitude improvement when using genetic calculator method TA2 relative to CPI.
We also assessed relative efficacy when the victim profile was known, and just one unknown contributor was inferred. The average log(LR) match information reported on eight adjudicated CLR cases was 13 (10 trillion). The average genetic calculator TA1 match information on these same cases was 18 (quintillion), a five order of magnitude improvement. Thus, for both one and two unknown contributors, the genetic calculator mixture interpretation method is more informative than the CPI and CLR match statistics.
Reproducibility was measured on these 16 mixture cases by obtaining duplicate computer solutions for each case. The average match information deviation between the two independent solutions was under half a log(LR) unit.
We conclude from this study that a genetic calculator can provide reliable mixture interpretation. Specifically, when inferring either one or two unknown contributor genotypes, the genetic calculator is effective relative to current methods. Moreover, we quantified reproducibility using match information. This genetic calculator has already been admitted into evidence in a Frye jurisdiction. Our validation study (for efficacy and reproducibility) establishes the genetic calculator's reliability under the additional prongs of Daubert.