Forensic Scientist Reveals Disturbing Error Rates in DNA Database Matches

The National Offender database, which contains the DNA profiles of offenders, has more than 13 million profiles, almost double since 2009. However, there is growing concern about the accuracy of DNA matches and the potential for human error in the system. Forensic scientist and professor Dan Krane revealed alarming error rates in DNA test results on the "Hidden Killers" podcast with Tony Brueski.
 
 Krane, who has extensive experience working with DNA databases, expressed his excitement about the potential to identify error rates or determine errors in DNA test results. He cited his involvement in an investigation for the state of Victoria in Australia, where he had access to about 15,000 DNA profiles. "When you did all possible pairwise comparisons of those 15,000 people, what you found is that there were a fair number that matched perfectly at 14 out of 18," Krane explained. "There were fewer that matched at 15 out of 18, but still some, fewer that matched at 16 out of 18, but still some, and then a whole bunch that matched at 17 out of 18, more than matched at 14 out of 18, and then a whole bunch that matched at 18 out of 18."
 
 Krane explained that the only explanation for the ones that matched at 18 out of 18 was that these were people who were entered into the system twice, possibly under an alias. However, the ones that matched at 17 out of 18 indicated that there was a mistake in one of the two times their DNA got entered into the system. This translated into an error rate of one in 300. "And when you're talking about chances of a coincidental match in the quintillions or the bazillions, if you prefer, yet you make a mistake every 300th time that you try to generate a DNA profile, the bazillion number doesn't matter anymore. It gets swapped out by that chance of making a mistake," Krane said.
 
 This error rate is particularly concerning as it means that opportunities to generate useful investigative leads are missed. If a DNA sample is collected from a crime scene and there is no suspect, authorities search the CODIS database to find a match. However, every 300th time this is done, a lead is not generated, which Krane believes is an unacceptably high failure rate given the consequences of having a rapist or a murderer go without being apprehended.
 
 Krane suggested that one way to correct this could be by doing a pairwise comparison of the 13 million profiles in the federal database to determine if errors happen more frequently at one of the locations than others. "Maybe one of those loci just is more error-prone than others. Maybe the biology just doesn't work as well for that one locus. That one location. And when you do your database search, maybe you don't include that one, right? Or maybe you include it and this, but you factor in that you discount it because that one is just more prone to having a mistake," Krane suggested.
 
 Krane also expressed concern about the secrecy surrounding the CODIS database and suggested that more people in the correct positions should have access to it to identify more criminals and solve more cases. He also mentioned the development of probabilistic genotyping systems, expert systems used to interpret test results that human analysts find too complicated. These systems, while not openly shared due to stringent non-disclosure agreements, can distinguish between signal and noise, which is particularly useful when dealing with a small amount of DNA or a sample with many different contributors.
 
 While the expert systems may not explicitly use artificial intelligence (AI), they involve serious computer science undertakings with hundreds of thousands of lines of computer code. These systems may help reduce human error and improve the accuracy of DNA database searches, ultimately helping to catch more criminals and solve more cases.
 
 Krane's insights highlight the need for more transparency and improvement in the system t