Back to Newsroom
Corrections to NIST Mixture Interpretation Webcast Misstatements about TrueAllele Casework
On Friday, April 12, 2013, the National Institute of Standards and Technology (NIST) hosted a free one-day workshop on interpreting forensic DNA mixtures in casework. This workshop was webcast live to maximize participation by forensic DNA analysts.
The workshop underscored the limitations of human DNA mixture interpretation, and highlighted computer methods that represent genotype uncertainty using probability. As in their previous workshops, NIST described Cybergenetics TrueAllele® Casework, a widely used probabilistic genotyping method.
There were errors in the NIST lecture on probabilistic genotyping. NIST's description of the TrueAllele technology contained both mathematical and logical mistakes. Cybergenetics promptly notified NIST (April 15), and enumerated NIST's misstatements to them in a corrections letter (April 17). NIST published some clarifications (April 19), but they have not yet corrected all their errors.
In NIST's mixture example, TrueAllele had accurately placed over half of its probability on the true genotype. However, NIST confused likelihood and probability. By getting the math wrong, NIST underreported TrueAllele's probability four-fold.
Since TrueAllele correctly placed 56% of the genotype probability, it preserved 56% of the DNA match statistic and reported 10.3 (18.5 was the maximum). NIST compared TrueAllele's result with output from another computer program that used a different population database. By switching to an Oceania population, NIST made an abundant American genotype seem rare, and so overstated the other match statistic ten-fold.
The TrueAllele probabilistic genotyping method is easy to understand, extensively validated and widely used in criminal cases. The method is thorough (considers all possibilities) and objective (does not know the suspect's genotype). TrueAllele solves DNA mixture problems that human analysts cannot. The Cybergenetics casework system preserves DNA identification information that helps solve crimes and prevent victimization.
The corrections and clarifications that Cybergenetics provided NIST (at their request) are listed below. The slide numbers are from NIST's probabilistic genotyping lecture.
- TrueAllele got it right
- TrueAllele genotype likelihood
- TrueAllele genotype probability
- TrueAllele likelihood ratio
- TrueAllele genotype listing
- New Zealand population database
- New Zealand genotype probability
- New Zealand likelihood ratio
- TrueAllele validation studies
- TrueAllele admissibility
- TrueAllele reports and trials
(Slide 28) TrueAllele® Casework objectively infers genotypes from STR data, thoroughly considering all allele pair possibilities. TrueAllele separates mixtures into contributor genotypes using all the peak height data, without being told the suspect's genotype. The computer's inferred genotype is independent of anyone's preferred answer. With data ambiguity, genotype probability is placed on multiple allele pairs, one of which might match a suspect or a known contributor.
(Slide 29) The table shows a prior population genotype (HWE) and TrueAllele's computer-inferred evidence likelihood values (Prob).
(Slide 30) The genotype probability is proportional to the product of prior times likelihood (HWE*Pr). Dividing each product by the Total sum 0.0143 normalizes the column to a posterior genotype probability. The genotype probability at the suspect's FGA allele pair 20,22 is HWE*Pr/Total, or 0.0080/0.0143, which equals 56%.
The likelihood ratio (LR) is computed at the suspect's row (FGA allele pair 20, 22) by dividing the posterior genotype probability 56% by the prior genotype probability 5.4%, to give a LR of 10.3. The maximum possible LR (e.g., for a single source sample) at this locus would be 100%/5.4%, or 18.5. A genotype probability for the suspect of over half thus preserves most of the identification information.
The genotype listing shows 9 allele pair rows at a 99.99% cumulative probability level. In the TrueAllele visual user interface (VUIer™), the user controls this probability level, typically set around 95%. A 99% level would show 6 rows, a 95% level only 5 rows, and a 90% level just 4 rows.
(Slide 34) The STRmix operator used a New Zealand native (Ma̅ori) enriched population database. Therefore, the LR values in the two examples cannot be directly compared.
STRmix inferred an FGA genotype probability of 24% at allele pair 20,22.
Using a US Caucasian population database, the LR at 20,22 would be 24%/5.4%, or 4.4. The very large LR shown (103) results from the relative rarity of FGA genotype 20,22 in New Zealand (very small population denominator), and cannot be meaningfully compared with TrueAllele's US Caucasian statistic (much larger 20,22 population prevalence).
TrueAllele DNA mixture interpretation has been extensively validated. Three peer-reviewed validation papers have been published, one using NIST laboratory data of known composition (PLoS ONE, 2009) and two using casework items (JFS, 2011 & 2013). Other mixture validation studies have been presented at conferences (ISFG, 2011; AAFS, 2013).
TrueAllele Casework has had successful admissibility hearings in Pennsylvania, California and the United Kingdom. TrueAllele has statewide precedent in Pennsylvania, based on favorable appellate rulings.
Over 100 criminal TrueAllele case reports have been written. TrueAllele testimony has been given in 16 criminal trials. Both the prosecution and defense use TrueAllele to determine accurate LRs from complex DNA mixture evidence.
Perlin MW, Sinelnikov A (2009) An information gap in DNA evidence interpretation. PLoS ONE 4: e8327.
Perlin MW, Legler MM, Spencer CE, Smith JL, Allan WP, et al. (2011) Validating TrueAllele® DNA mixture interpretation. J Forensic Sci 56: 1430-1447.
Coble MD, Butler JM (2011) Exploring the capabilities of mixture interpretation using TrueAllele software. International Society of Forensic Genetics. Vienna.
Perlin MW, Dormer K, Hornyak J, Schiermeier-Wood L, Greenspoon S. Virginia TrueAllele® validation study: casework comparison (A125) ; 2013; Washingon, DC. American Academy of Forensic Sciences. pp. 94.
Perlin MW, Belrose JL, Duceman BW (2013) New York State TrueAllele® Casework validation study. J Forensic Sci 58: in press.