State v. Simmer: The First High Court Opinion on "Probabilistic Genotyping"

Last week, the Nebrasks Supreme Court affirmed the admission of testimony about what has been dubbed "probabilistic genotyping." Twelve years ago, in 2007, Joy Blanchard was discovered on the floor of her home. Two knives protruded from her neck. A spindle broken from the banister lay nearby. The cause of death was blunt force trauma and stab wounds to the head and neck. Investigators swabbed the spindle, the handles on both knives, and the interior front door knob. Nine years passed before DNA testing indicated the presence of DNA from Joy's nephew, Charles Simmer, on one of the knife handles and the door knob.

Charged with first-degree murder Simmer filed a pretrial motion to preclude “any and all testimony concerning DNA testing and the results of said testing,” including “identification” and “comparison.” After an evidentiary hearing, the district court denied the motion. At the jury trial, the state presented both "probabilistic genotyping" from Dr. Mark Perlin, the inventor of a complex software program named TrueAllele, that "the match between the knife handle and Simmer was 3.71 thousand times more probable than a coincidental match to an unrelated person, while the match between the doorknob and Simmer was 4.22 quintillion times more probable than a coincidental match to an unrelated person." Simmer was convicted and sentenced to life imprisonment.

What made the "genotyping" "probabilistic" is not that genes were typed. The STRs used in the identification are not genes. They are repeated, noncoding sequences of DNA base pairs (such as TAGATAGATAGATAGATAGATAGATAGA) that are sprinkled within and between genes. Neither Is the typing "probabilistic" because probabilities were quoted. Ordinary DNA typing results are accompanied by "random match probabilities."

Rather, unlike ordinary DNA profiling of a single-source samples, these probabilities were not derived from any one individual's features as inferred by a human examiner. The samples were mixtures of DNA from (at least) two men, and no one could say which two individual profiles were in the mixture. The computer program, however, could estimate the probability of the mixed-up collection of DNA features displaying the measured "peak heights" when coming from DNA from several individuals including the defendant as opposed to the probability of the same data coming from DNA from individuals not including the defendant.

Dr. Perlin testified to the ratio of those probabilities (more technically, of those probability densities) using the confusing phrase "match probability" when in fact there was no simple match. He framed the question as "How much more does the suspect match the evidence at this location than a random person?" But any random profile either would match -- in the sense of having the same alleles as those giving rise tom the observed peaks -- or it would not. Because of the ambiguity as to the true profiles in the mixture, the common-sense notion of a match -- of inclusions to and exclusions from -- the class of possible contributors to the mixture is not part of the logic of "probabilistic geneotyping."

Be that as it may, the issue on appeal in State v. Simmer, 304 Neb. 369 (2019), was not how to present the output of the software. It was whether that output could be the subject of any testimony at all. The Nebraska Supreme Court ran through the usual "Daubert factors." Despite arguments about the type of validation studies, the court was impressed with the sheer number of studies and the length of time that the system has been used.

The defendant's expert was Nathaniel Adams, who "has a bachelor’s degree in computer science and is employed by a company that 'consult[s] with lawyers.'" Mr. Adams testified (correctly) that one cannot directly validate the accuracy of particular likelihood ratios and that the software had not been developed and tested from the standpoint of software engineering, which, he suggested, requires independent study of source code over long periods of time. The Supreme Court was not impressed. It held that validation "from a software engineering perspective" was not essential and that the publication of research on the performance of the software (as best as that could be measured for known mixtures and via simulations) was sufficient.

The defense also maintained that "the validation studies must be discounted because Perlin is a coauthor of some of the publications and, as the owner of the company that owns. TrueAllele, has a financial interest in seeing it found reliable" To this, the Supreme Court responded that there were plenty of studies in which Perlin was not an author, but it also suggested that the source of validation studies is irrelevant because that fact pertains only to credibility:

In other words, Simmer is contending that there are reasons to question Perlin’s credibility. Even assuming that is true, an attack on the credibility of an expert witness is not aimed at the expert’s reasoning or methodology and is thus not the proper basis for a Daubert/Schafersman challenge. See Smith v. Colorado Organ Recovery Sys., 269 Neb. 578, 694 N.W.2d 610 (2005) (stating that challenge to expert witnesses' credibility is not Daubert/Schafersman claim).

This cannot be correct. Although the credibility of a particular witness certainly is fair game at trial, Independent replication of experiments is an important factor in scientific practice. The absence of validation from independent researchers should count against admissibility, although it would not necessarily be dispositive under Daubert.

Another questionable feature of the opinion lies in its discussion of the PCAST Report. In 2016 and 2017, the President's Council of Advisors on Science and Technology evinced dissatisfaction with the extent to which probabilisitc genotyping as applied in some situations had been validated. The software developers disputed the report -- but later responded with additional studies. The peculiarity in Simmer is not that the court determined that the output of TrueAllele was admissible despite PCAST's reservations. It is that the Nebraska Supreme Court refused to read the PCAST Report because it "is not in [the] record [on appeal]." Instead, it looked only at an obscure and incomplete "Washington trial court order containing a description of the report." On that basis, it wrote that "with no scientific evidence in the record indicating that TrueAllele could not generate reliable results under the circumstances here, we cannot say that a conclusion that TrueAllele could reliably be applied to the facts of this case would amount to an abuse of discretion."

But why was appellant was "unable to cite directly to the report"? It is a duly published report of a prestigious advisory body. Appellate lawyers and courts frequently cite books, articles, and reports about scientific research and policy. National Academy reports and Office of Technology Assessment reports have long been cited and discussed in opinions on the admissibility of DNA evidence. They address "legislative facts." Legislative facts apply across cases." In contrast, "adjudicative facts" are not general truth; they are the how, what, where, and when of specific cases.

Under the Nebraska Rule of Evidence (see, e.g., Neb. R. Evid. 201(a)), only adjudicative facts need to be proved at trial. Both appellate and trial courts have the power to consult published scientific opinions on the limitations of a scientific method. Here, the alleged limitation is that the software was not shown to work well with mixtures that contain very small fractions of a suspect's DNA. Courts should be more open to receiving information in the scientific and related literature on such issues.