All mixed up: Responding to flawed opposition case reports on complex DNA evidence

W. Allan, J. Bracamontes, K. Danser, M. Legler, M. W. Perlin, "All mixed up: Responding to flawed opposition case reports on complex DNA evidence", Promega's Thirty Fourth International Symposium on Human Identification, Denver, CO, 21-Sep-2023.

Poster

Download Poster

Abstract

Low-level DNA mixtures can be important forensic evidence. In a recent federal case, a California man was accused of drug possession. The defense lawyer requested DNA testing. Cybergenetics’ TrueAllele® system found strong exclusionary likelihood ratios (LR) of a millionth. STRmix™ software found weak LR values near one tenth. The prosecutor retained opposition expert William Thompson, who used the numerical discrepancy to harshly criticize TrueAllele. Who was right?

Fully Bayesian TrueAllele examined all the DNA mixture data to find reproducible match statistics. Probabilistic genotyping (PG) STRmix software only saw about 10% of that data, insufficient for a conclusive answer. Both programs were right, based on different methods and data. Thompson’s case report raised many irrelevant claims, but it ignored the data amounts and betrayed little understanding of PG methods. His report was completely wrong.

The scientific portion of this talk corrects Thompson’s mischaracterizations, presenting the PG methods accurately. A STRmix sensitivity study of peak height thresholds shows how different data inputs yield LR values spanning 12 orders of magnitude.

The response portion builds on this science to explain why Thompson is mistaken. The talk examines many of his conceptual errors, drawing from a rebuttal of 120 assertions.

Talk highlights include:

TrueAllele and STRmix probabilistic genotyping software both analyzed the same low-level two-person mixtures at a defendant’s request.
Both programs found exclusionary results, with Cybergenetics reporting TrueAllele LR values around one millionth, and a lab reporting STRmix LRs near one tenth.
TrueAllele used all data at 21 STR loci, while STRmix was limited to 24 peaks over 14 loci, a density of 1.7 peaks per locus, insufficient for informative mixture analysis.
A sensitivity study on varying data thresholds found STRmix LR values ranging over 12 orders of magnitude, far greater than any difference with TrueAllele.
Thompson’s case report was unfounded and inaccurate. 120 of his statements warranted response.

Defendants are entitled to meaningful DNA evidence. With low-level mixtures, entering more data and updating more variables can potentially deliver more exculpatory LR information. Thompson advised crime labs to “punt” when they can’t interpret DNA. But using more effective software, they can go for the goal of truth.

Links

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