Nationwide, forensic DNA labs have quietly been encountering troubling results when testing the newest computerized DNA analysis methods. Complex DNA mixtures give analysts real difficulty. To help with this, law enforcement recently began replacing human analysts with a computer program for some portions of that analysis. The software—generically called probabilistic genotyping—claims to reduce analytical error. Some labs are discovering, however, that the software can get a critical portion of cases wrong—badly wrong.
Take the Los Angeles County Sheriff’s Department’s Scientific Services Bureau. In their internal validation study of a probabilistic genotyping system, the Bureau created experiments testing DNA mixtures made up of first-order relatives. First-order relatives include an individual's parents, full siblings, and children. They did this specifically to figure out “how well the software can distinguish between closely-related individuals.” The results were abysmal.
Almost 90% of tested mixtures made up of 3 people or more produced results that falsely included a person the testers knew was not included in the mixture. In other words, the false positive error rate was 87%.
And those erroneous conclusions were not insignificant. Even in these false positive cases, the system claimed high confidence in its result. The Bureau warned: “Analysts should use caution when interpreting mixtures believed to be comprised of first-order relatives.”
And LA County is not alone. We’ve found eight other labs that encountered the same thing. On a scale where any value greater than 1 is inclusionary and larger magnitudes indicate elevating confidence levels, all of these labs discovered troubling false positives when relatives are involved: