Peeling away uncertainty: a probabilistic approach to DNA mixture deconvolution

W. Allan, H. Chaudhry, J. Donahue, M. W. Perlin, M. Wilson, "Peeling away uncertainty: a probabilistic approach to DNA mixture deconvolution", American Academy of Forensic Sciences 74th Annual Scientific Conference, Seattle, WA, 25-Feb-2022.

Talk

PowerPoint presentation of the conference talk.

Download Handout
Download PowerPoint

Abstract

After attending this presentation, attendees will understand how genotype conditioning improves DNA mixture separation, decreases genotype uncertainty, and increases DNA match information.

This presentation impacts the forensic science community by describing how the genotype conditioning method increases the identification information that can be obtained from complex DNA mixtures.

Complex mixtures contain DNA from multiple contributors in differing amounts. Unlike older manual methods, probabilistic genotyping (PG) can use all of the peak data to separate the DNA contributors out of these mixtures into their component genotypes. Once the genotypes are inferred, they are compared to reference genotypes (relative to population allele frequencies) to calculate likelihood ratio (LR) match statistics. These statistics can be inclusionary or exclusionary.

Genotype comparison LR values scale with inferred genotype certainty, where more definite genotypes show higher probability allele pair possibilities. More definite genotypes lead to larger (i.e., more informative) match statistics, whether inclusionary or exclusionary.

Uncertainty in the inferred genotypes – seen as a diffuse allele pair probability distribution – reduces this match information. Genotype uncertainty can be due to incomplete separation of the DNA contributors, low DNA amounts, or biological relatedness between contributors. More genotype uncertainty leads to less LR information. Therefore, identification information can be increased by reducing genotype uncertainty. Genotype conditioning, or ‘peeling’, assumes a statistically established contributor genotype in genotype inference. This conditioning ‘locks down’ a contributor genotype, focusing the inference on the other unknown contributor genotypes. Conditioning can improve the genotype mixture separation, reducing uncertainty in the other genotypes, and increasing match information.

This study demonstrates the utility of genotype conditioning using the TrueAllele® PG system. Mixture samples (genotypes and their amounts) were randomly designed. Two, three, four and five DNA contributors were examined. Initial computer processing established the most informative DNA contributors. Subsequent computer runs used these statistically established genotypes to reduce uncertainty in the remaining unknown genotypes. Multiple rounds of processing ‘peeled away’ the unknown contributor genotypes until one remained. Assuming all contributor genotypes as known gave highly accurate contributor mixture weights.

Inferred genotypes were compared with 70 references to calculate LR match statistics. The log(LR) information increased linearly with contributor DNA amount, until reaching a maximal information plateau for major contributors. Genotype conditioning improved separation for overlapping contributors, overall doubling their match information. For example, a five-contributor mixture contained two 11% components. Initially, the LRs for these overlapping contributors were in the thousands. Peeling these contributors away from the major contributors increased their LR values into the billions, while peeling the contributors away from each other increased the LRs to almost a quintillion.

An adjudicated quintuple homicide case demonstrated how genotype conditioning works to support criminal justice. Prosecutors used a probative five-person mixture sample, taken from a spot of blood on a garage floor, to establish victim family members’ presence at a suspect’s home. The forensic question was to determine which family members left their DNA in the blood spot. While initial computer results were informative, more analysis could be done.

Through multiple rounds of genotype conditioning, TrueAllele separated the garage floor mixture into component family member genotypes. Peeling away DNA contributors statistically showed the presence of both parents, their children, and another relative. Conditioning also showed that another family member, whose DNA was not in the evidence, was statistically excluded from the mixture. While family allele sharing had initially indicated an inclusionary match statistic, genotype peeling factored away genotypes of relatives to show her DNA was not in the mixture.

This presentation demonstrates both the laboratory and real-world utility of genotype conditioning, and how the method reduces DNA mixture uncertainty by increasing information.

Links

American Academy of Forensic Sciences 74th Annual Scientific Conference - Program