Back to Presentations
Objective DNA Mixture Information in the Courtroom: Relevance, Reliability and Acceptance
M.W. Perlin. "Objective DNA Mixture Information in the Courtroom: Relevance, Reliability and Acceptance", International Symposium on Forensic Science Error Management: Detection, Measurement and Mitigation, Arlington, VA, 22-July-2015.
Hosted by the National Institute of Standards and Technology (NIST).
PowerPoint presentation with audio recording of Dr. Perlin's narration.
DNA mixtures arise when two or more people contribute their DNA to a biological sample. Mixtures are seen in sexual assault kits, homicide evidence, handguns and other "touch DNA" surfaces. With advances in detection technology, they have become the predominant form of DNA evidence in many crime laboratories. While DNA from one person is easy to interpret, mixture data has complex patterns comprising many allele peaks of varying height.
One person's DNA produces either one allele peak, or two of similar height, so a height "threshold" is meaningful. But data-simplifying thresholds fail to give accurate results when applied to complex mixture patterns. Ten years ago, NIST demonstrated a ten order-of-magnitude match statistic discrepancy between crime laboratories analyzing the same mixture data. Mixture "inclusion" analysis tests whether a subject's alleles are included in a set of (thresholded peak) alleles, but is inherently subjective – the analyst sees the subject's genotype during the analysis.
An entirely objective (and potentially more informative) approach is to first separate out the genotypes of each mixture contributor without ever seeing the subject, and only afterwards make a comparison. This can be accomplished by sophisticated computing that considers many thousands of genotype alternatives, and how well their linear combinations explain the quantitative data. Multiple possibilities for a contributor genotype are assigned probabilities. Faithful modeling of the laboratory process can yield genotypes that accurately preserve DNA identification information.
Comparison of a separated evidence genotype with a subject's reference genotype, relative to a population, gives a match statistic. This statistic is a simple ratio – the probability of genotype match divided by the random match probability. It is also a likelihood ratio (LR), or Bayes factor (BF), a standard measure of information change based on observed evidence. The LR is mathematically probative because it assesses how evidence data affects a hypothesis (i.e., whether the subject contributed their DNA to the mixture). And the LR's assessment is nonprejudicial, because (as a BF) the ratio factors out prior belief about the hypothesis. Thus genotype separation addresses FRE 403 relevancy balancing.
The reliability of objective genotype separation has been extensively tested for at least one such system. Dozens of independent and developmental validation studies have been conducted, with seven peer-reviewed publications. These studies use the LR as an objective information measure to assess the method's sensitivity (true positives), specificity (false positives) and reproducibility (close numbers). This extensive testing, error rate determination, and scientific peer-review address FRE 702 and Daubert reliability factors.
Courts have accepted this extensively validated computer method, which has withstood Daubert and Frye challenges in six states. Admissibility has been upheld at the appellate level. Separated genotypes provide easy to understand results.
Objective DNA analysis elicits identification information from evidence. Validation establishes accuracy and error rates. Courts require solid science – extensively tested and empirically proven – to promote criminal justice, societal safety, and conviction integrity.