Back to Presentations
Assessing TrueAllele® genotype identification on DNA mixtures containing up to five unknown contributors
M.W. Perlin, J. Hornyak, R. Dickover, G. Sugimoto, and K. Miller, "Assessing TrueAllele® Genotype Identification on DNA Mixtures Containing up to Five Unknown Contributors", American Academy of Forensic Sciences 66th Annual Meeting, Seattle, WA, 20-Feb-2014.
After attending this presentation, attendees will understand the applicability and limitations of genotype modeling solutions for DNA mixture problems. This presentation will show the conditions under which genotype modeling can compute a reliable DNA identification.
The presentation will impact the forensic community by establishing the generality of genotype modeling for DNA interpretation, and validating the use of Cybergenetics TrueAllele® Casework system on mixtures having many unknown contributors.
Manual review of complex DNA evidence does not fully elicit all the data's identification information. Therefore, computer methods have been developed for mathematical interpretation of mixed and low-template DNA. The genotype modeling approach computationally separates out the contributors to a mixture, with uncertainty represented through probability. Comparison of a contributor genotype to another genotype, relative to a population, calculates a likelihood ratio (LR). Validating an interpretation method on a broad range of DNA mixtures having known composition can help predict an expected LR outcome in a particular case.
This randomized experimental design examined 40 DNA mixture items. The 4 mixture sets had 2, 3, 4 or 5 contributors, with each item specified as a random mixture weighting of randomly assigned known references. Both normal (1 ng) and low (200 pg) template amounts were studied, for a total of 8 groups (4 contributor numbers x 2 template amounts) each having 10 mixture items.
The mixture weight (MW) of each item's contributors had a predetermined design value, but was subject to laboratory variation. For each item, the TrueAllele system computed two MW estimates, one using all the known genotypes, and the other with all genotypes unknown. MW was also computed manually on the 2 contributor items. There was a strong association (r2 = 0.999) between the three computed MWs for an item, and less (r2 = 0.907) with the design value (p < 10-12). The computed TrueAllele known-genotype MWs had the most precise values (average sd = 0.0195 log(LR) units), and were used in the remainder of the study.
Following a procedure used in a previous validation study1, scatterplots were developed comparing a contributor's known DNA quantity (logarithm of MW x total DNA, x-axis) versus its identification information (log of LR, y-axis). This approach permitted examination of all the match results (all contributors of all items) within their groups across a single statistical analysis. The scatterplots of positive match results were roughly linear (r2 = 0.638), showing expected log(LR) reductions for equal MWs and high DNA amounts. The average regression slope was 12.66 log(LR)/log(DNA) (p < 10-40), so a ten-fold change in DNA amount yielded a trillion-fold change in LR.
Analysis of covariance (ANCOVA) of the eight groups showed different x-intercept values, but no significant difference in slope (p = 0.348 > 0.05). This slope invariance was observed across four different contributor numbers (2, 3, 4 and 5) and DNA template amounts (200 pg and 1 ng). This invariance indicates that TrueAllele's information response to DNA mixture data is relatively independent of contributor number or template amount. The ANCOVA outcome suggests that this genotype modeling method produces reliable match results, regardless of the DNA mixture composition.
The false exclusion rate (Type II error) was estimated as a function of MW. For normal DNA amounts, there were positive match results in 100% of comparisons (0.10 ≤ MW ≤ 1.00), 82% (0. 05 ≤ MW ≤ 0.10), 40% (0.01 ≤ MW ≤ 0.05) and none below 0.01. With low-template DNA, positive match results were found in 100% of comparisons (0.25 ≤ MW ≤ 1.00), 91% (0.10 ≤ MW ≤ 0.25), 24% (0.05 ≤ MW ≤ 0.10) and none below 0.05. In addition to these sensitivity and specificity results, reproducibility was measured in all groups.
This validation study used randomly generated DNA mixtures (reflective of actual casework samples) of up to 5 contributors, with both high and low template amounts, to assess TrueAllele genotype modeling. The study found that the computer's MW values were reliable, and that match information changed with DNA quantity in a predictable way that did not significantly depend on contributor number or template amount. Type II error was determined as a function of MW. This in-depth experimental study and statistical analysis show the applicability and limitations of the TrueAllele method.