A single swab of the thigh of a college student who had been raped and murdered was the crux of a murder case in Texas up until last week.

The one right thigh swab of 18-year-old Haruka Weiser, taken shortly after her violent April 2016 death, yielded an extraction of DNA originally found to exclude the suspect.

The swab underwent a second extraction procedure at the request of the state. It resulted in a low-level mixture which at first, through manual interpretation, determined it was inconclusive. The second interpretation, using an automated software program used to sort out mixtures, yielded a few more genetic markers—and the prosecution’s case focus in tighter on Meechaiel Criner, the accused. By that second look at the second swab, the likelihood ratio determined it was 192,000 times more likely the genetic mixture was Weiser and Criner than Weiser and any unknown individual out in society.

But the prosecutor’s DNA evidence—using STRmix—was tossed by a judge last week. The judge ruled that the Texas Department of Public Safety forensic analyst who did the manual (Combined Probability of Inclusion, or CPI) and automated STRmix procedures did not follow the protocols for the use of the software.

The instance marks at least the second such time that STRmix DNA findings were tossed from a high-profile murder case, the first being in an upstate New York murder case in 2016.

However, the debate over the merits of probabilistic genotyping—machine sorting of complex DNA likelihoods—remains for future courtroom battles, according to some experts consulted by Forensic Magazine about the latest case.

“All of these software developers say that it’s an objective system, and it deconvolutes mixtures that could never be done before,” said Allan Jamieson, of the Forensic Institute, a defense expert on the Criner case. “Well, it’s not really objective, because somebody has to choose which peaks are going to be plugged into the machine. This case exemplified that, because they actually ran it first, got a number that didn’t suit them, and so they changed the parameters and ran it again—and got a number that suited them better.”

A spokesman for STRmix, which originated in New Zealand, said in a statement that the debate was not with the program’s validity—instead it was about the protocols used in manipulating the tool.

“The validity of STRmix was never argued in court,” the statement reads. “Judge Wahlberg is quoted as stating, ‘It turns out we don’t get to those issues in this case at this time. In short, the issue was not with STRmix.'”


Weiser, an Oregon woman who was just finishing her first year at the University of Texas at Austin, was found strangled with a ligature and sexually assaulted in an area near Waller Creek.

The swab of her thigh became one of the featured pieces of evidence in the prosecutor’s case, which was in pre-trial hearings last week.

But the former DNA section manager working for the Texas DPS last week told a courtroom that she could not produce a required portion of her handwritten notes taken during DNA analysis on the case, she said.

Judge David Wahlberg found that the procedural problems meant the DNA mixture evidence produced by STRmix could thus not be used at trial—thus eliminating the majority of genetic evidence against Criner.

All that remains is some mitochondrial DNA data left to be admitted, according to accounts.

Prosecutors have said they would not challenge the judge’s decision to exclude the STRmix analysis—but had enough evidence to proceed with their case.

Criner’s murder trial is expected to begin July 9.


The defense’s experts had prepared a voluminous defense against the state’s STRmix evidence before it was tossed. Such defense could be brought to bear in cases to come, according to the experts. 

Jamieson, the defense expert, was focused on the likelihood ratios. He had prepared in excess of 150 slides in a PowerPoint show to help the courtroom understand the Criner mixture conundrum.

Primarily, determining parameters for software like STRmix means the results can be changed, from system-run to system-run.

“The fact that they ran it presumably using the parameters that the validation had told them to use, and got a poor number that didn’t support the case,” said Jamieson. “To then change those parameters, perhaps just to see if it did improve their case, that is … naughty, at best.”

But there are other issues, he said. During the amplification process, DNA can have stutter. Such stutter, or interference, can appear like different peaks, showing genetic markers that may not be there, as he told Forensic Magazine in an interview this week.

“The whole issue of stutter is problematic … A tiny little peak could be from a suspect—or an artifact,” said Jamieson. “It’s very difficult to ascertain which it is.”

The problem is where to establish thresholds, separating true genetic markers (alleles) from the ripple of interference showing “ghost” markers that could point a jury in a misleading direction, Jamieson said.

In the New York murder case from two years ago, a judge ruled that such issues in interpretation were “unduly prejudicial to the defendant” Oral Nicholas Hillary. The judge ruled that STRmix creator and developer John Buckleton, an expert for the prosecutor, had found the witness would not be allowed to “pick and choose data from different ‘reliable sources.’”

Suzanna Ryan, a California DNA expert involved in high-profile cases across the country, was tapped by the Criner defense to handle some cross-examination questioning, as well as non-mixture issues.

She told Forensic Magazine that she wanted to see the arguments for and against STRmix’s use play out in the courtroom—since she remains unconvinced either way.

For instance, one sample in this Criner case had seven alleles foreign to the victim, but STRmix determined an inconclusive result. But the second extraction of the same sample had only four foreign alleles—but it resulted in the 192,000 likelihood ratio.

The judge heard hours of STRmix and DNA mixture testimony, Ryan said—and while it appears the rejection of the evidence is based on procedural aspects by the analyst, questions about the use of the software remain.

“I was looking forward to hearing the opposing sides to the STRmix arguments as I am still sort of on the fence about STRmix,” she said. “If it works as it is alleged to, then I think it could be a useful tool for the forensic DNA community. However, there are some facets to STRmix analysis that do not make sense to me, which make me question the results.”

STRmix, again, determined that their tool is working fine—it’s just the operators who may present a problem.

“It is important to recognize that the issue with the DNA in this case was not with STRmix … the forensic scientist in the case did not follow the laboratory’s standard operating procedure,” according to STRmix.

STRmix also has a competitor called TrueAllele, made by a company based out of Pittsburgh. While it was not used in the Criner case, it was used on the same sample in the upstate New York case—and distanced the suspect from the mixture under the victim’s fingernails.

Jamieson said all the computer programs need to be further tested, to understand their strengths and limitations.

“These software systems may work, but I think we’ve done enough cases now to say that our sufficient strange results (can) say, let’s pause, and get these tested by people other than the developers, on real casework,” he said.