Skip to main content

DNA software used in Caneiro murders 'not ready for prime time,' engineer testifies


Six-minute read

play
Show Caption

FREEHOLD - As the Public Defender's Office forged ahead in its battle to keep DNA evidence derived from a cutting-edge computer program out of court in the quadruple murder case against Paul Caneiro, a software engineer testified Tuesday there is inadequate documentation that the computer program was tested to ensure it works properly.

Mats Neimdahl, head of the Department of Computer Science and Engineering at the University of Minnesota, said at an ongoing pretrial hearing that he found little or no documentation the STRmix computer software program that analyzed DNA evidence in the case against Caneiro was adequately tested to ensure reliable results.

Christopher Godin of the Public Defender's Office asked Neimdahl if, upon review of the documents, he formed an opinion on the likelihood there are undetected flaws in the STRmix software. 

"Well, given the lack of documentation, it's hard to form any opinion about this in terms of the quality of the software, which leads me to say, 'No, this is not ready for prime time,'" Niemdahl responded.

The hearing is before Superior Court Judge Marc C. Lemieux, who will decide whether the DNA evidence against Caneiro derived from the STRmix software will be admissible at the 57-year-old Ocean Township man's upcoming trial in the 2018 murders of his brother, Keith, 50; sister-in-law Jennifer, 45; niece Sophia, 8; and nephew Jesse, 11.

The Pubic Defender's Office is seeking a ruling invalidating evidence derived from the software, asserting the method it uses to analyze DNA is not generally acceptable in the scientific community. 

The challenge represents the first time a court in New Jersey will rule on the validity of the software, which uses a method known as probabilistic genotyping to analyze DNA, deviating from traditional methods of DNA analysis that have been in use for decades.

Lemieux's decision will have statewide implications, as the the New Jersey State Police recently began using STRmix in its DNA lab, while the Public Defender's Office is seeking to block its use in criminal cases going forward.  

The ruling almost certainly will be appealed to the state Supreme Court, prosecutors and defense attorneys on the Caneiro case have said.

Emergency workers responding to a slow-burning fire at Keith Caneiro's mansion on Nov. 21, 2018, discovered the bodies of the victims. Keith Caneiro had been shot four times in the head and once in the back. Jennifer Caneiro and the two children were repeatedly stabbed and badly burned. Jennifer Caneiro also was shot in the head.

Prosecutors allege Paul Caneiro committed the murders after his brother discovered he was stealing from the companies they co-owned. They allege he then set fire to the Colts Neck mansion to cover up the crimes before returning to Ocean Township and setting fire to his own home to mislead investigators and make it look like the family was being targeted by violent thugs.

DNA that the STRmix program determined to be a mixture from Paul and Sophia Caneiro, and another mixture to which Jesse was deemed to be a contributor, was later found on clothing recovered from the defendant's basement.

STRmix and other probabilistic genotyping programs use algorithms to analyze small amounts and complicated mixtures of DNA that often can't be analyzed by traditional methods.

The traditional method, known as random match probability, generates a statistic on the probability a match to a DNA profile can be found in the general population. 

Probabilistic genotyping generates a 'likelihood ratio" that a person of interest can either be included or excluded as a contributor to a DNA mixture.

Niemdahl testified the STRmix software, as it is used in criminal prosecutions, is deemed to be "critical safety software," like software used in medical devices and aircraft and, as such, it is crucial to validate that the software works as it is intended. 

But, of STRmix, he said, "There is no real documentation of any actual tests that had been run, nor where those tests might have come from.

"We have no idea what faults, critical failures, could be lurking in there," Niemdahl said of the STRmix software.

"You would expect an organization to be able to pull all of this out to show me how they've done things," he said. "There was very, very little documentation of any testing that was done."

Niemdahl said the resulting danger is that the program will produce erroneous results, and the DNA analysts won't realize the results are wrong.

He discussed two possible errors — that a person who did not contribute to a DNA mixture was identified as someone who had, and a person who was a contributor was identified as not having contributed to the mixture.

If a result is reported as "a near certain and that's being accepted in court, it might lead to false convictions or at least, seriously biasing the jury," Niemdahl said. 

"I can see similar things happening the other way," he said. "I would think that if the software reports that the DNA is not part of a mixture, that would be good for the defendant."

Niemdahl said the primary users of probabilistic genotyping are prosecutors and law enforcement, who are not keen to see it fail. 

"From the prosecutorial side, the incentive structure is set up to succeed in convictions, so pointing out flaws with the labs may not be in the best interest, the best setup, for detecting and reporting anomalies," he said.

Christopher Decker, deputy first assistant Monmouth County prosecutor, pointed out during his cross-examination of the witness that more than 80 laboratories in the United States are using STRmix, and each had to undergo validation studies of the software to gain accreditation from the FBI. 

As part of those validation studies, "there have been hundreds of thousands of tests that have been published by STRmix," Decker said. 

"What has been published is hundreds of thousands of runs of STRmix, but I haven't seen any publications about why those specific tests were selected, what those specific tests were, what parts of the code those tests exercised, and so on and so forth," Niemdahl said. 

Kathleen Hopkins, a reporter in New Jersey since 1985, covers crime, court cases, legal issues and just about every major murder trial to hit Monmouth and Ocean counties. Contact her at khopkins@app.com.

close