Defeating opposition experts using science in a recent case example

J. M. Bracamontes, W. P. Allan, M. W. Perlin, "Defeating opposition experts using science in a recent case example", Promega's Thirty Fifth International Symposium on Human Identification, San Antonio, TX, 24-Sep-2024.

Poster

Download Poster

Abstract

In 2020, a man was shot following an altercation. He later died from his injuries. Video surveillance linked a suspect to the murder. There were no witnesses to the shooting. Police collected the gun and magazine used in the crime. Was the suspect the killer? Could DNA evidence help?

The local crime lab tested the firearm items for DNA. Their DNA data showed at least two (gun) or three (magazine) contributors to the mixtures. A private DNA lab interpreted the data and compared it with the defendant’s DNA profile. By manual inspection, the lab excluded the defendant from the major DNA contributor. However, the lab couldn’t draw any conclusions regarding the minor contributors, for either evidence item.

When the suspect’s defense attorneys received the inconclusive DNA results, they reached out to Cybergenetics for “probabilistic” genotyping computer interpretation. Objective TrueAllele^® Casework analysis statistically excluded the defendant as a DNA contributor. Separated genotype components included a 6% minor contributor (gun), and a 2% minor contributor (magazine). Cybergenetics reported exclusionary likelihood ratio (LR) match statistics, along with exact false exclusion error rates (ER)¹.

The government retained an expert to review the TrueAllele results. The opposition expert’s report contained many flawed arguments. A notable flaw was an incorrect description of computed error rates. Relying on this opposition report, the government filed a Daubert motion to challenge TrueAllele admissibility.

The opposition expert argued that TrueAllele had a high binary false exclusion error rate (LR<1 for true contributors). For minor contributors, they described a 60% error rate for the 1-5% mixture range, and an 18% error rate for the 5-10% range. They based their claim on a published validation study² from a previous TrueAllele version. (In the version used in the case, these hypothesized error rates would be lower at 35% and 0%, respectively.) Where was the flaw in their argument?

Error rate depends on LR¹. That mathematical fact is given in the error rate law ER ≤ LR – exclusionary error rate can never exceed the likelihood ratio. The error rate is only meaningful relative to LR, giving the chance that other people would be excluded as strongly. But the opposing expert ignored error rate’s dependence on LR.

Cybergenetics calculated¹ and reported relevant error rates in the case. However, the government incorrectly applied a cutoff of 1 for a binary error rate, discarding the LR value. That is not how to determine forensic DNA error rates. The relevant context is how strongly the evidence matches (or doesn’t match) the suspect – the actual LR statistical support for the suspect, not some cherry-picked cutoff level.

We revisited the validation paper’s LR data. We showed that the opposition’s purported “error rates” entailed weak exclusionary LR values near 1 (Figure 6, red crosses) from less informative genotypes. These validation points were not relevant to the case’s highly informative genotypes, which gave strong exclusionary LR values (Figure 6, green line) of one in 70 million (gun) or 160 billion (magazine). The prosecution’s spurious argument was entirely unrelated to TrueAllele reliability or its results in the case.

Responding to the Daubert challenge, Cybergenetics prepared a 26-page declaration. We detailed the opposition report’s flaws, refuting inapplicable arguments with science and facts. We described the TrueAllele technology’s error rate, admissibility, and court acceptance. Based just on document submissions alone, without even an admissibility hearing, the judge admitted the TrueAllele results as reliable scientific evidence.

References

1. Perlin, M.W. Efficient construction of match strength distributions for uncertain multi-locus genotypes. Heliyon, 4(10):e00824, 2018.

2. Perlin M.W., Hornyak J., Sugimoto G., Miller K. TrueAllele^® genotype identification on DNA mixtures containing up to five unknown contributors. J Forensic Sci. 2015;60(4):857-868.

Links

Promega's Thirty Fifth International Symposium on Human Identification - Site